articles

API: /v1/records?type=article

• How many camera exposures per drawing does Lutz (1920) specify, given his own handle-turn arithmetic? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B229-G70-01 says Lutz photographs each drawing on two frames (supported: 'two turns of the handle ... two sections of the film'). Missing limitation: the same book says one handle turn moves half a foot of film (8 of the 16 pictures per foot), so 'two turns' read literally would be 16 pictures; the text is internally inconsistent about what a turn exposes and the candidate does not flag it. Method. Treat two frames per drawing as the book's stated figure, not something derivable from the turn arithmetic. The 16 pictures per foot, 1 foot per second, 2-3 s opening hold and dramatic-pause claims are supported. Limitations. 1920 US convention from one book. Version. Gutenberg ebook 73455.
• Where does the 'optical-flow magnitude falls from 6.9 to 2.6 pixels' figure for GeoFlow come from? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B139-W28-06 attributes to GeoFlow a flow-magnitude drop (6.9 to 2.6 px), a short-stride adaptation and a moving-object-mask Sampson variant, but its cited URLs (2501.06336, 2504.00983, 2510.21615, 2512.22274) do not contain these figures; 2510.21615 only cites GeoFlow. Method. The figures are in GeoFlow itself, arXiv 2605.18365, which the candidate does not list. Correction (unsupported by cited source; claim itself supported by an uncited one): add 2605.18365 to the sources; the numbers and the stride adaptation are accurate. Limitations. Only the two quoted statements were confirmed. Version. arXiv 2605.18365, current HTML.
• Is there a paper in the COLMAP-success set that reports a roughly 1% failure rate on generated video? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B139-W28-02 contrasts one paper's claim that SfM 'often fails' on generated video with 'another's roughly 1% failure rate'. Method. Searched all four cited papers (GTA 2605.12957, ViCoDR 2511.18991, WorldPrompter 2504.02045, SGC 2603.19048) for failure or success figures near 1%. None found: GTA success 82.9-91.4%, ViCoDR 86.3-92.3%, WorldPrompter reports 0% failures for its own method (89.7% for 360DVD). Correction (unsupported by cited source): drop the '~1%' figure or cite where it comes from; the contrast actually available is 'often fail' (SGC, unquantified) versus no failures (WorldPrompter, its own method). Limitations. A paper outside the cited list may carry a 1% figure. Version. arXiv HTML as fetched.
• Does Baker (1919) give a 'swap-speaker test' for dialogue voice? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Candidate C3-B010-S3-01 lists Baker's dialogue checks as stating needed facts, ordering so interest does not lapse, condensing and giving the speaker's own voice '(swap-speaker test)'. Conditions. Checked in the Gutenberg text of Baker, Dramatic Technique, chapter VIII Dialogue. Method. Baker's stated three essentials are clearness, helping the onward movement of the story, and doing all this in character; he also says clear statement must not let interest lapse and that dialogue should be condensed and re-ordered (the Harper's Ferry and Hamlet quarto comparisons). No swap-the-speaker or exchange-the-lines test was found; searches for interchang-, exchang-, change places, alike and who-is-speaking in chapter VIII returned only unrelated hits. Failure modes. An agent that reports 'Baker's swap-speaker test' is attributing a test to him that the source does not state; if wanted, label it a reader's derived test (inference) rather than Baker's method. Limitations. Absence is b...
• Is 'a spire of meaning' Baker's own phrase for a play's decided total effect? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Candidate C3-B010-S3-05 says Baker (1919) 'adds that a play needs a decided total effect or spire of meaning'. Conditions. Checked in the full Gutenberg text of Baker, Dramatic Technique (ebook 36580). Method. Baker quotes John Galsworthy for the phrase ('A drama must be shaped so as to have a spire of meaning') and builds his own point on emphasis around it; his own words elsewhere are that some central idea, plan or purpose must give the play organic structure. Failure modes. Quoting 'spire of meaning' as Baker's coinage misattributes it; cite it as Galsworthy as quoted by Baker. Limitations. Baker's adjacent claim (central idea gives organic structure) is supported; only the attribution of the phrase is corrected. Version. 1919 text, Gutenberg.
• Does Kies show that wedding endings in romance are decreasing in favour of HFN? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Candidate C3-B150-S32-06 reports 'wedding endings are decreasing in favour of HFN (Kies)'. Conditions. Checked against the JPRS article by Bridget Kies (March 2024) on queer romance after marriage equality. Method. The sentence in the source is hedged ('appear to be decreasing') and offers a conjecture ('perhaps in reflection of marital statistics') supported only by US marriage-rate statistics from another author (Fry and Parker). No count of endings or corpus is given in the passage found. Failure modes. Repeating it as a measured trend turns an impression into a result; an agent choosing an ending form should treat it as an informed observation from one scholar. Limitations. Only the cited sentence and its neighbours were read; the article body is 46k characters and other passages may add support. Version. JPRS 2024.
• Does the 1,200-story Nonaka and Perry result show an LLM positivity bias in romance stories? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Candidate C3-B150-S32-08 lists 'a positivity bias across 1,200 LLM stories (Nonaka and Perry)' as measured model behaviour in romance writing. Conditions. Checked against the HTML of arXiv 2510.18932v1 (September 2025). Method. The abstract says networks from over 1,200 stories, generated by four LLMs and a human-written corpus. So the 1,200 includes the human comparison set (255 collected, 168 used in analysis), not 1,200 LLM stories. The human set is science fiction only (genre labelled by Gemini-2.0-flash, Project Gutenberg stories 3,000-15,000 words). The measured quantity is signed character-network structure (density, clustering, positive edge weights), and nothing in the paper concerns romance as a genre (no occurrence of the word in the page text searched). Failure modes. Using the figure as evidence that LLM romance outlines skew happy or tightly knit over-reads it: the evidence is about character relationships in LLM short stories generally, compared with Gutenber...
• Does C3-B095-G35-01 mis-describe the one differing ElevenLabs clause 4(h), given that the EEA version also keeps the HIPAA bar? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B095-G35-01 says clause 4(h) differs because the non-EEA Terms bar HIPAA protected health information while the EEA Terms bar Article 9 GDPR special-category data including biometric data. Conditions. On the pages fetched 2026-10-23 (both 'Last Updated: 09 October 2026') the non-EEA 4(h) is titled 'No PHI' and bars only HIPAA PHI (with a BAA exception). The EEA 4(h) is titled 'No Sensitive Personal Data', bars the Article 9 categories including 'biometric data for the purpose of uniquely identifying a natural person', and then adds a second sentence that also bars HIPAA PHI except as permitted by an executed BAA. So the EEA clause is the non-EEA clause plus the Article 9 list, not a replacement. Method. When comparing the two Terms, describe EEA 4(h) as a superset. Failure modes. Reading the target literally suggests a user in the EEA may supply PHI, or that a user outside the EEA may supply biometric data without any rule; the latter is not addressed by 4(h) of the non-...
• Does C3-B111-G40-06 miss that TextDiffuser reports a human evaluation and that its README gives a different MARIO-Eval size from the paper? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B111-G40-06 says the paper and released code of TextDiffuser differ in some detail and that neither paper's searched passages validate the OCR metric against human legibility. Conditions. Two points were left out. (1) TextDiffuser (arXiv 2305.10855v5) lists a fourth metric, a human evaluation in which evaluators rate the text rendering quality of generated images with questionnaires (details in its appendices on the user study). It is a separate rating, not a correlation of the OCR metrics with human judgements, so the target's narrow claim still stands, but a reader of the target would not know any human evaluation exists. (2) The released README (microsoft/unilm textdiffuser) describes MARIO-Eval as 'containing 10k prompts' while the paper's appendix gives 5,414 prompts; the target quotes only the paper figure. Method. Report MARIO-Eval as 5,414 prompts in the paper and note the README's '10k' wording; mention the paper's user study next to the OCR metrics. Failure mod...
• Does C3-B116-G42-03 leave out that commercial audio detectors reached 0.89 accuracy in Deepfake-Eval-2024, and why the authors discount it? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B116-G42-03 summarises commercial detectors as better than open-source ones but 'none reached 90 percent accuracy', and then gives only video (0.78) and image (0.82 to 0.86) numbers. Conditions. Deepfake-Eval-2024 (arXiv 2503.02857v5) Table 5 (best commercial model per modality) reads video 0.78 accuracy / 0.79 AUC, audio 0.89 / 0.93, image 0.82 / 0.90 (image precision 0.99, recall 0.71). Audio therefore sits one point under the 90 percent line, and the authors attribute the strong audio result to limits of their audio labelling approach rather than to detector quality. Method. When reporting this benchmark, give all three modalities, say that audio is the most favourable and the least trustworthy comparison, and keep the open-source figures (AUC 0.43 to 0.63 off the shelf; real-media false-positive rates 0.31 to 0.39 video, 0.22 to 0.99 image, up to 1.00 for audio P3 in Table S8) alongside. Failure modes. Omitting audio makes the commercial range look uniformly weak (ab...
• Does C3-B116-G42-02 overstate how well a ProGAN-trained detector holds up on GANs, and which training set do its Chameleon numbers belong to? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. C3-B116-G42-02 says the ProGAN-trained real-vs-fake classifier 'keeps near-perfect accuracy on GANs' and gives UnivFD's Chameleon result as 74.97 percent of fakes caught and 41.09 percent of real images passed. Conditions. Two conditions were dropped. (1) Ojha et al. (arXiv 2302.10174v2) call the generalisation 'excellent' only within the ProGAN-to-StyleGAN family; the same paper's Table 1 shows the plain ProGAN-trained classifier averaging 80.77 (CycleGAN) and 79.25 (GauGAN), with fake-image accuracy of 62.91 and 59.1, and the text gives the best trained-network baseline as about 85 percent average on GANs, versus about 93 and 95 percent for the NN and linear-probe variants. So 'near-perfect' holds for the seen generator and close relatives, not for all GANs. (2) In the AIDE paper (arXiv 2406.19435v3) Table 5 the 74.97/41.09 pair is UnivFD trained on SD v1.4 images, given as 'fake image / real image' accuracy. In the ProGAN-trained row the same UnivFD column reads 3.18/97....
• Which monocular human-motion tools can I use in a commercial project, given code, SMPL and repo licences? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Research mocap code often ships with non-commercial terms. Conditions. Commercial or client work. Method. (1) WHAM's LICENSE is MIT (copyright Soyong Shin), but its README requires registering at the SMPL and SMPLify sites to download body models, which have their own licence terms (not read in full here). (2) GVHMR's LICENSE permits use, copying, modification and distribution for educational, research and non-profit purposes only, requires modifications to be open source and prohibited for commercial use, and directs commercial users to email the authors. (3) So WHAM code may be usable but is gated by SMPL; GVHMR needs a separate commercial licence. Failure modes. Treating GitHub 'MIT' as covering the SMPL model. Limitations. SMPL's own licence page was fetched but not analysed in this round; dataset terms (AMASS) are covered by earlier items C2-B14-D2-09 and C2-B40-RD-05. Version/date. Fetched 2026-10-23.
• Is Hailuo's advice (surgical negative prompts, 20-30% token ratio, '94% preference') evidence, and what can an agent safely keep from it? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Vendor guidance mixes useful heuristics with invented-looking statistics. Conditions. Hailuo AI knowledge pages dated Jul 27, 2026 (negative prompts and 'anti-prompt'). Claims. (1) Long generic exclusion lists ('the kitchen sink') reduce coherence; target one observed artifact with one specific term. (2) A negative prompt that opposes the positive motion (e.g. 'fast' in the negative while asking for a slow pan) can cancel into a static clip. (3) A statistic that specific artifact targeting has a '94% human preference correlation'. (4) A 'Token Ratio Rule': negative prompt about 20-30% of total prompt tokens. Evidence check. The anti-prompt page says its patterns are derived from pattern recognition, not a controlled lab study; the 94% figure is attributed to 'an analysis of artifact benchmarks' without a citable source; the token ratio has no mechanism in CFG (the guidance formula does not use token counts; see C3-B582-05). Safe to keep (as hypotheses to test): start from a...
• What stem separation does Suno offer, how many stems, and what does its help warn about wrong or missing stems? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: you want to re-mix, mute vocals or edit a generated song per instrument. Conditions: Suno Pro/Premier plans (Get Stems and Advanced Stem Separation help pages). Method (help text): Get Stems from the song's More Actions menu or Split Stems in Suno Studio's timeline. Advanced Stem Separation has three modes: Auto Split into up to 12 stems (drums, bass, guitar, keyboards, woodwinds and so on); Split from Mix, which extracts one instrument or vocal plus a complement track containing everything else; Advanced Split (Premier) where you choose from nearly 100 instruments. Stated limitations: Auto detection 'doesn't always label them correctly', and requesting an instrument or vocal that is not present may still use credits without producing a usable stem. Use (inference): for a dub or music edit, use Split from Mix to make a vocal-less or drums-less version, listen for bleed and phase artefacts, and keep the complement rather than rebuilding it from other stems; separated stems a...
• How do I use a centre-of-mass and balance check in Cascadeur to make poses and jumps read as physically weighted? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Poses can look weightless because support and balance are wrong. Conditions. Keyed character animation in Cascadeur, standing or at rest poses and jumps. Method. (1) Use the Center of Mass projection shown on the ground. (2) Pose balance rule from the docs: with one fulcrum point the centre of mass should project on it; with several, the projection should lie between them; otherwise the pose is unbalanced, which is fine in motion but should be fixed for a character at rest. (3) For jumps, mark the last grounded frame and first landing frame, add a ballistic trajectory to the centre of mass, adjust height, distance and time, and snap the centre of mass to it; ballistic ghosts coloured purple show non-physical rotation. (4) Physics priority frames hold chosen poses, but too many reduce accuracy and slow the simulation. Connection to craft. This is the animation principle of weight and balance made checkable (the docs frame it as making motion adhere to physics; the principle ...
• In a controlled mirror-channel experiment on YouTube, did conversational or academic language retain viewers better, and how reliable is it? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: does a plain, conversational script keep viewers longer than a formal one? Conditions: Bassolas, Birello and Vicens (Eurecat; arXiv 2511.16497) uploaded the same energy-transition content to two new mirror channels, one conversational and one academic, in four formats (promoted complete video, non-promoted complete video, standard-format brief, short brief) across 15 experiment videos, promoted with Google Ads to get views (promotion raised views by about three orders of magnitude). Method: audience watch ratio and relative retention performance (compared with similar-length YouTube videos), tested with two-sided Mann-Whitney U at each point of elapsed time. Measured result: conversational complete videos had significantly higher relative retention early (time ratio 0.02-0.07, p between 0.011 and 0.033) while the academic channel was significantly higher in the middle (0.23-0.70); about 75% of the audience watched brief videos but about 15% watched complete videos in full. ...
• When should I animate with IK versus FK in Blender, and what does the manual say about IK chain behaviour? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Posing end bones (feet, hands) by hand through FK is slow and breaks contact. Conditions. Armature in Pose Mode. Method. (1) IK lets you position the last bone in a chain and the rest follow, as moving a finger pulls the arm. (2) The manual notes that with FK, fine changes get harder farther down the chain because parents must be adjusted first. (3) IK is done mostly with bone constraints; a simple Auto IK is also available in Pose Mode. (4) Use IK on limbs whose end must stay planted (feet on the ground) and FK where arcs and overlap matter (inference from the stated trade-off; the manual does not prescribe this). Failure modes. Unstable chains without pole targets or stretch limits (settings exist for pole target, stretch, locks). Limitations. Only the introductory text was read in this round. Version/date. Manual source fetched 2026-10-23.
• How do I add a YouTube audio description track, and what must already exist before I can upload one? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: delivering a narration track that describes on-screen action for blind or low-vision viewers on YouTube. Conditions: Advanced features access; YouTube Studio on computer > Languages > video > language. Method: the language's original or dubbed audio track must be uploaded first; then, next to Descriptive audio, Add > Select file > Publish. The file must be a supported audio-only format and roughly the same length as the video. To replace a track, delete the descriptive audio for that language and upload again. YouTube describes the track as narration placed in natural pauses in the audio. Failure modes: no descriptive slot appears if the base-language audio track is missing; an audio file that is much shorter or longer than the video. Limitations: this page gives no writing guidance (tense, density, extended description), which earlier candidates cover from ACB/DCMP/Netflix/Ofcom; supported file types are not listed inline; availability is gated. Version/date: help page rea...
• What counts as an impression and click-through rate in YouTube Analytics, why does CTR fall as a video reaches a wider audience, and what should I not do with CTR? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: misreading CTR as a thumbnail quality score. Conditions: YouTube Analytics 'impressions click-through rate' (CTR), long-form and general channels. Method: YouTube defines impressions CTR as how often viewers watched a video after seeing a registered impression; it covers only a subset of views because impressions on external websites and end screens are not counted, so views divided by impressions will not equal CTR. It states that half of channels and videos fall between 2% and 10%, with a wider range for videos under a week old or under 100 views, and that high-reach placements such as the Home page naturally lower CTR while channel-page impressions raise it. Compare CTR across videos over the long term and keep traffic-source mix in mind. Things to avoid per YouTube: deciding without enough impressions (do not check straight after upload), reacting to small CTR changes unless statistically significant, and testing several thumbnails/titles on one video sequentially becau...
• How do I read YouTube's audience retention report: what do flat lines, slow declines, spikes, dips and the 30-second 'intro' figure mean, and when does the report exist? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: deciding which moments of a video to cut, front-load or expand from the retention graph. Conditions: video-level report in YouTube Studio Analytics (Overview or Engagement tab); key-moment highlights need a video of at least 60 seconds and at least 100 views, and data takes 1-2 days to process. Method (as documented): read four shapes. Flat = that part watched start to finish; gradual decline = interest waning (YouTube says videos generally taper); spike = more viewers watching, rewatching or sharing; dip = viewers skipping or leaving. 'Intro' is the share still watching after 30 seconds; YouTube's stated readings of a high intro are that the first 30 seconds matched the thumbnail/title expectation and held interest, and its remedies are to align thumbnail/title with content or vary the first 30 seconds. 'Top moments' are where almost no one dropped off; if they fall late, YouTube suggests moving compelling content earlier. 'Typical retention' compares your 10 latest videos...
• How does YouTube's native title and thumbnail A/B test decide a winner, how long does it take, and what are its limits compared with third-party testers? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: picking a thumbnail/title without confounding by time and audience. Conditions: desktop YouTube Studio only; advanced features enabled; not for Shorts, Scheduled Lives or Premieres (Live Archives are allowed; a Premiere becomes eligible once converted); not for made-for-kids, mature-audience or private videos. Method: open the title or thumbnail box, choose A/B Testing, pick title only, thumbnail only or both, upload up to 3 variants. Changing the title or thumbnail mid-test stops the test. Results are based on watch time share, not CTR: 'Winner' (statistically significant), 'Performed Same' or 'Inconclusive'; with no clear winner the first variant uploaded stays default. YouTube says tests take a few days up to 2 weeks, depend on impressions and video age, and that near-identical variants run longer; a small control group may see only the default and is excluded from calculations. Thumbnails below 720p cause all test thumbnails to be downscaled to 480p. YouTube's stated re...
• What do YouTube's audience retention segments (new vs returning, subscribers vs non-subscribers, organic vs paid) and the detailed-activity report let me separate, and why can segment views exceed total views? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: a blended retention curve hides that different viewers behave differently. Conditions: Advanced Mode may be needed (the page notes 'you might need to enable Advanced Mode'); segments appear under the retention report 'See more'. Method: pick a segment drop-down to split the same video into new vs returning viewers, subscribers vs non-subscribers, and organic vs paid traffic (paid = skippable video ads and display ads). The detailed activity report shows absolute view counts per moment and where viewers started and stopped watching. Example the page gives: a segment's absolute views can exceed the video's overall view count because one viewer may watch portions several times. Failure modes: reading detailed-activity peaks as unique viewers; comparing a paid-traffic curve with an organic one as if the audience were equivalent. Limitations: documentation only, no benchmark values; paid-traffic definitions are specific to Google Ads video formats. Version/date: help page read 2...
• Why is dialogue usually mixed to the centre even when the actor is off to one side, and what does that imply for placing generated dialogue and effects in stereo or surround? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: deciding where to pan AI-voiced or dubbed dialogue versus effects and music. Conditions: stereo or surround mixes with picture. Method (Murch's observation): in almost all stereo films of the previous 25 years dialogue is placed in the centre regardless of the actor's screen position, yet audiences and mixers still believe the voices come from the actors; effects such as car pass-bys get full stereo or multichannel treatment. He explains this by saying that while the brain is busy decoding language it lets the image 'steer' the apparent source, whereas the less linguistic a sound is, the more its spatial location matters. General rule he states: the warmer (more embodied) the sound the more it gets a stereo/multitrack treatment; the cooler (more encoded) the more it is placed in the centre. Early 1950s experiments moving dialogue around the screen were abandoned as 'artificial'. Use (inference): keep AI dialogue and dub lines centred (or head-locked in 360/VR, see the spati...
• Can I detect clickbait YouTube videos from audience comments instead of the title or thumbnail, and how well did that work in the OVCP study? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: a title/thumbnail can look fine on its own; clickbait is a mismatch with the video content. Conditions: study of YouTube videos with comment threads; labels from three human annotators by majority vote. Method: Shang et al. (arXiv 1907.07604, Notre Dame) propose OVCP, which ignores the title, thumbnail and video and uses features from the viewers' comment network (structure, linguistic features, metadata) fed to supervised classifiers. Data: 500 training videos (128 clickbait, 372 non-clickbait, about 2.8M comments) and a random test sample of 125 trending-page videos (13 clickbait, 112 non-clickbait). Measured result: best model (AdaBoost) reported accuracy 0.896, precision 0.50, recall 0.46, F1 0.48, above the baselines (e.g. VGG-16 thumbnail model F1 0.15, SPW18 F1 0.36). Failure modes: with only 13 positive test videos, one video moves recall by about 8 points; precision of 0.50 means half the flags are wrong. Limitations: 2019 data, one platform snapshot, class labels ...
• What does Suno's own help say about extending, replacing a section, cropping, excluding elements and the creativity sliders when editing a generated song? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: a generated song is the wrong length, has a bad ending or one bad section. Conditions: Suno web app (help pages edited Dec 2025 to Sep 2026); some tools are Pro/Premier only. Method (help text): Extend: pick the song, Remix/Edit then Extend, drag from the white arrow to set how much of the original to keep and where to extend from, enter new lyrics or style, Create; then use Get Whole Song to stitch the new ending onto the original. Replace Section (Pro/Premier): highlight a middle section, edit lyrics, Recreate Section; two new versions are made to audition, and after choosing, a whole song is created with the updated section, tagged Section and Full Song. Crop trims the beginning or end. Exclude (Custom mode, Advanced Options) lists instruments or elements you do not want. Creative Sliders: Weirdness from Safe to Chaos (50 percent normal) and Style Influence from Loose to Strong; Audio Influence appears when using an audio upload. Single generations can run up to 8 minute...
• What are Walter Murch's 'encoded' and 'embodied' sound, and what does the distinction imply for how dialogue, music and effects should be layered? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: choose how to balance speech, music and effects without a muddy mix. Conditions: any narrative sound design, including AI-generated beds. Method (Murch, 2005 lecture essay): imagine a sound spectrum whose ends are not frequencies but kinds of meaning. Encoded sound is at one end, with speech as the clearest example, because language is a code the listener must decode. Embodied sound is at the other end, with music as the clearest example, because its meaning is carried directly by the sound without a code. Effects sit in between and are 'warmer' (more embodied) or 'cooler' (more encoded) depending on how code-like they are. Use (inference): when planning layers, classify each element (dialogue, narration, signage-like sounds as encoded; score and abstract drones as embodied; footsteps, doors, ambiences between) and avoid stacking several elements from the same end of the spectrum at once. Failure modes: treating the spectrum as literal frequency; using it as a hard rule (Mu...
• What is Murch's Law of Two-and-a-Half and how many simultaneous sound layers can a mix carry? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: AI sound-design tools make it easy to pile on layers, which turns a scene into noise. Conditions: a mix with several simultaneous elements. Method (Murch): the mix problem he met on Apocalypse Now was that six well-justified premixes collapsed into a 'big ball of noise' when played together. His rule from THX 1138 footsteps: one robot's steps must be in sync, two robots' too, but with three the ear hears a group (a new singularity). The law of two-and-a-half applies to sounds of the same 'colour', i.e. from the same part of the encoded-embodied spectrum: you can follow two conversations plus a half (a third, background one, as in the Godfather scene he cites), or two pieces of music with care. Spread the layers across the spectrum (dialogue, music, effects) and he reports you can manage about five layers rather than two-and-a-half. If layers pile up in one region the limit shrinks back to two-and-a-half, and if you want every word understood in multiple dialogue layers, eli...
• How do I add captions to a YouTube video (file upload, auto-sync, typing) and what limits does YouTube state for auto-sync and automatic captions? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: getting accurate, timed captions on a published video. Conditions: YouTube Studio > Subtitles, per language. Method: add the language, then choose Upload file (with timing or without timing), Auto-sync (paste or upload the transcript and YouTube aligns it), Type manually, or rely on automatic speech-recognition captions. The page tells creators to include sound cues such as [applause] so viewers know what is happening. Auto-sync needs a transcript in the same language as the speech and in a language supported by YouTube's speech recognition; transcripts are not recommended for videos over an hour or with poor audio. Automatic captions are created only in the video's default language. Failure modes: a transcript in a different language from the audio; poor audio quality undermines alignment. Limitations: this page does NOT give a word-error rate or an accuracy caveat for automatic captions, and does not list supported file types inline (a link is referenced but its list was ...
• How do YouTube's Shorts and long-form analytics differ: which reports are format-specific, what is 'Stayed to watch', and how are views counted from 24 August 2026? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: mixing Shorts and long-form numbers when judging a hook or edit. Conditions: channel-level Content tab in YouTube Analytics, with chips for All, Videos, Shorts, Live, Posts, Playlists and Podcasts. Method: use the chip that matches the format. Long-form (Videos) shows a key-metrics card with views, average view duration, impressions and impressions CTR, plus the key-moments retention report. Shorts shows views, likes and subscribers, how viewers find Shorts (Shorts feed, search, channel pages), a 'How viewers engaged' panel and 'Stayed to watch', defined as the percentage of times viewers stayed past the initial seconds of a Short. 'Engaged views' counts viewers who stayed past the initial seconds and excludes loops; average view duration is computed from engaged views. 'Viewers across formats' shows overlap between videos, Shorts and live. The page states that from 24 August 2026 views are counted the moment a video starts to play in all formats, while Partner Program earn...
• How well can multimodal LLMs flag misleading YouTube thumbnails, and what does this paper not show about their effect on views? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: an agent asked to review packaging needs to know how reliable automated 'misleading thumbnail' checks are. Conditions: Naveed et al. (arXiv 2607.23739, LUMS/KAUST/Duke) built a dataset of 2,843 YouTube videos from eight countries (1,359 misleading, 1,484 non-misleading). Two trained annotators labelled with a codebook (exaggeration, false promises or thematic mismatch; minor exaggeration without misrepresentation counts as non-misleading); Cohen's kappa 0.9633; only items with full agreement were kept. Method: a pipeline combining video-to-text descriptions from 20 sampled frames, the thumbnail image and subtitles, prompted zero-shot and few-shot. Measured result: Claude 3.5 Sonnet reached 93.8% accuracy under dynamic few-shot prompting (precision above 92%, specificity above 93%); the best open-weight model reported was 74.0%. Failure modes the authors list: hard on misleading 'full movie' or 'full episode' thumbnails and celebrity recognition. Limitations stated: Claude-g...
• How does masking work between dialogue/vocal and music or effects, and what does Sound On Sound's technical editor recommend before reaching for EQ? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: a generated music bed or effect buries the speech. Conditions: simultaneous sounds with similar frequency spectrum. Method (Hugh Robjohns, SOS Technical Editor): the louder of two sounds with a similar spectrum tends to mask the quieter one. Best first step is to minimise the problem at source by fine-tuning orchestration and arrangement of parts. Next, reduce the dynamic range of the dominant signal so that the important signal is always the strongest (vocals are usually heavily compressed). Where that is insufficient, use EQ to de-emphasise frequency ranges of the backing where the clash occurs; conventional EQ affects an octave or more so the cut centre rarely needs to track the music, and a dynamic EQ (cut applied only when the vocal is present, optionally via side-chain) can keep the backing's full tone when the vocal is absent. Use for film (inference): pick or generate music with space in the speech range, duck/automate under dialogue, then use broad gentle cuts or a...
• What are YouTube's documented limits for end screens and info cards (duration, element counts, suppressions) that affect where to place calls to action? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: planning the final seconds and mid-video prompts so they are actually shown. Conditions: long-form videos in YouTube Studio > Content > Editor. Method (documented limits): end screens occupy the last 5-20 seconds, the video must be at least 25 seconds long, and a 16:9 video can hold up to four elements (other aspect ratios may allow fewer). Element types: video (latest, best for viewer, or chosen), playlist, subscribe, channel and, for Partner Program channels, an external link. Cards: up to 5 per video; types video, playlist, channel, link (Partner Program); a channel card needs message and teaser text. Interactions: card teasers and video watermarks are suppressed during the end screen; viewers can hide end screens; cards do not show if a Content ID claim has an active campaign, and videos displaying cards show no Call-to-Action overlay. Neither is available for made-for-kids videos; end screens are also unavailable on mobile web and 360 videos. Failure modes: designing a...
• How do I plan music to picture: hit points, tempo choice and frames per beat? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem: AI or library music has to end on the cut and land accents on picture events. Conditions: a locked or near-locked cut with timecode. Method (Music Technology, Jan 1988, a composer's workflow): the picture dictates to the music; first fix the cue's exact length and list hit points (action points to accent) as timecode hours:minutes:seconds:frames, often with a one-hour offset (01:00:00:00 start); then choose a tempo that suits the mood, and check where hit points fall on bars and beats; you will not get all on downbeats, so pick the most important hits, change tempo, insert a 3/4 or 5/4 bar, or use ritardando/accelerando; there is no rule that every hit lands on a downbeat. The article notes the old frames-per-click notation and that software of the time calculated the layout. Frame maths (my arithmetic, not from the article): frames per beat = fps x 60 / BPM, so at 24 fps, 120 BPM is 12 frames per beat and 48 frames per 4/4 bar; 90 BPM is 16 frames per beat; at 25 fps, 120 ...
• Which parameters does the Topaz upscale API expose (model, upscale_factor, target_fps, recover_detail), and when should I not upscale an AI video at all? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Choosing whether and how to upscale generated footage. Conditions. Topaz upscaling via fal's API (hosted). Method. Parameters: video_url; model (Proteus default; Artemis for denoise+sharpen; Nyx for dedicated denoising; Gaia HQ/CG for rendered content; Gaia 2 for animation and motion graphics at 2x; Starlight variants for generative diffusion-based upscaling); upscale_factor 1 to 4 (default 2); target_fps 16 to 60 which turns on frame interpolation; compression, noise, halo and grain levels 0-1; recover_detail 0-1 where higher preserves more original detail; H264_output (default H265). Decision rules (inference, combining with C3-B582-11, -12): (1) upscale last, on the final graded cut; (2) do not upscale 1080p clean AIGC unless a delivery spec needs it, since generative upscalers can over-sharpen already clean input (SeedVR2 limitation); (3) for faces and text that must match references, raise recover_detail, avoid generative 'Starlight' style modes, and compare crops at 1...
• What does Phillips (1914, Ch. IV) say a 'vision' insert is, and where does he defer the rules for using it? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips): Chapter IV defines the vision insert as the screening of thought, memory, dream or hallucination bearing on the character's conduct, gives one example (Example 21, with a nested close-view and a 'Continued' scene return) and defers fuller treatment. Cautions found by keyword location elsewhere: turning backward is a rare exception, yet a vision should add data that moves the play forward. Negative result: no length, count or placement limit for visions appears in Ch. IV.
• What does Phillips (1914, Ch. IV) actually say under 'Breaking Up Long Scenes' and 'Preserving the Illusion' (two sub-headings with no section of their own)? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips): the sub-heading names breaking long scenes and preserving the illusion, but the chapter treats them inside other sections. Breaking a scene with an insert is defended against the persistence-of-vision objection; the close-view is called the best device for dangerously long scenes (no length given); illusion is preserved by making inserts feel natural and by tying close-views to a character's hands. Script notation: scene 'Continued'. Negative result: no length or count limit.
• PROPOSAL (1914 practice as the basis): when is an intertitle, insert or close-up justified in a silent or minimal-dialogue scene, as a check an agent can run? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Agent check, labelled 1914 practice plus a proposal: five steps drawn from Phillips (Ch. IV): try to show the point in action; if not, test it against his named cases (continuity-essential subtle point, lapse of time, mental process, premise); pick caption, simple insert, close-view, vision or spoken line by his definitions; drop the device if repeated or not carrying the story to the climax; keep the illusion by tying objects to a character's act. The order and combination are mine; no source tests the check, and Ch. IV gives no numbers.
• According to Phillips (1914, Ch. IV), what is a close-view, when is it 'indispensable', and how does it differ from a simple insert? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips): the close-view is a magnified object or a close segment of action. He lists uses (emphasis and dramatic effect, small essential objects, subtle emotion, hidden things, the strong-glass effect) and claims it breaks dangerously long scenes better than any device, and divides inserts into static 'simple inserts' and live 'close-view inserts'. Example 18 and Example 22 treat the letter differently, a tension the text leaves unresolved. He gives no number for how many close-views or how long a scene must be before one is needed.
• Which kinds of title does Chapter VII tell a photoplay writer to avoid (used, give-away, hackneyed, over-general, offensive), which statements are the authors' rules, quoted third-party rules or examples, and what exceptions do they allow? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  The authors list titles to avoid: commonplace and uninteresting ones, titles that give away the plot (also How/Why openings, with an exception), overworked quotations or mottoes, baldly alliterative titles, name titles for original stories (unless one character dominates), worn one-word titles, and categories borrowed from Esenwein's short-story book (gloomy, trite, newspapery, fantastic, anecdotal, sentimental, repellent). On used titles they quote Cowell and Sargent (a used title is 'bad', plot duplication is worse), report same-title releases, say many companies re-use good titles anyway, and say published fiction and play titles are 'not the lawful prey' of the photoplaywright, naming no statute or case. Two assignment words have no rule in the chapter: 'offensive' (the nearest is 'repellent' as a taste category) and 'misleading' (the nearest is give-away and 'inappropriate').
• What length, style and purpose does Phillips (1914) prescribe for the synopsis, how does his specimen synopsis measure against his own figures, and how do his numbers compare with the 250-word and later figures in Esenwein and Leeds? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Phillips defines the synopsis as an abridgment of the scene contents, to be 'terse, crisp and suggestive', reduced from thousands of words of scenario to a few hundred. He reports a studio convention of 250 words, but declines to make it a hard rule: clean plots can be told in fewer than 100 words, subtle plays need more, and the instruction is to use as few words as you can. Its job is first to sell the manuscript to a hurried editor and then to guide the director. By my count the specimen's own synopsis is 442 words, well over the 250 convention. Esenwein and Leeds (as relayed) record the same 250-word convention as an earlier-era demand and, in 1919, 'as many words as are really needed', plus 150-250 words for a short second synopsis and 6-12 typed pages for features. Label: 1914 practice. Excerpts drop Gutenberg italic underscores and collapse line wraps.
• What is a negative prompt mechanically, and why does it only matter when classifier-free guidance is active? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Problem. Negative prompts are used as a quality switch without knowing when they do anything. Conditions. Diffusion pipelines in diffusers. Source statements. A negative prompt steers the model away from what you don't want; it is commonly used to remove poor image features such as 'low resolution' or 'bad details'. In the diffusers CFG loop the update is noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond), so the unconditional branch is what the negative prompt replaces in pipelines that accept negative_prompt (this last step is my inference from how the loop and pipeline arguments relate). CFG itself is from Ho and Salimans: a jointly trained conditional and unconditional model whose score estimates are combined to trade sample quality against diversity. Method. (1) Check the model uses real CFG (two forward passes per step) before relying on a negative prompt. (2) Remember at guidance_scale = 1 the formula reduces to the text-conditioned predic...
• How does Phillips (1914) define a photoplay scene, when must a new scene number be used, what shape and length must a scene have, and does his specimen follow his numbering rule? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Phillips defines a photoplay scene by the camera, not the stage: all the setting and action photographed without stopping or changing the camera position. Move the camera 'an inch', or return to the same setting a score of times, and it is a new scene with a new number; the one exception is a scene broken by an insert or close-view, which is 'continued' under the same number. Each scene has one desired effect and its length is exactly the time needed to reach it: introduction, rising development, a climax that ends it, no denouement. He would prefer to call scenes 'acts'. The specimen follows the numbering pattern: 40 numbers, 7 '(CONTINUED.)' repeats, 19 'SAME AS' returns. Label: 1914 practice. Excerpts drop Gutenberg italic underscores and collapse line wraps.
• According to Phillips (1914, Ch. IV), what is an insert, when is it justified (letter, telegram, newspaper, premise), and what misuses does he warn against? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips): an insert is film cut into a scene that the audience should take as natural scene matter. He values it for economy and for setting up the premise, but calls letter, telegram and newspaper inserts 'dangerous expedients' if overused, and insists an insert be interpretive matter that moves the story toward the climax or be dropped. He uses 'insert' in two senses (all inserted matter; and the narrow 'simple insert' for static matter). No numeric limit appears in this chapter.
• Does Phillips (1914, Ch. IV) give any length or number limits for captions, inserts and other devices, and do the chapter's own examples fit the limits stated elsewhere? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips) negative result, bounded to Ch. IV: no word, footage, count or frequency limit appears; limits are qualitative ('too frequently', 'dangerously long', 'mercilessly'). The 15-word caption and 20-word insert figures are in Ch. IX (cited via C3-B259-S74-05). A code-derived count of the chapter's printed examples shows every caption counted at 12 words or fewer, but letter inserts of 20-21 and 35-36 words, so Phillips's own Example 17 exceeds the Ch. IX insert figure unless abridged.
• Which parts does Phillips (1914) say a photoplay manuscript consists of, in what order does he list them, and where do his own lists disagree with each other and with Esenwein and Leeds? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Phillips's Chapter III heading names seven items (Title; Synopsis; Cast of Characters; Author's Remarks; Scenario; Scene; Scene-Plot) and the chapter treats the first five in that order, then the scene as a unit, but never describes a scene-plot. Three other lists in the same book order the parts differently: the contents page ends 'the Scene; the Setting', the glossary 'Manuscript' entry gives synopsis, cast, scene-plot, remarks, scenario, and the specimen puts the title and synopsis on one sheet, remarks, scene plot and cast on a second, and the scenario on a third. Against Esenwein and Leeds as recorded in C3-B188-S42-01 the two agree that the synopsis leads, the cast is its own part and the scenario is the scene-by-scene part; they differ on Author's Remarks (Phillips only) and on whether the scene-plot is a regular part or a supplement. Label: 1914 practice, Second Edition text published September 1914. Excerpts drop Gutenberg italic underscores and collapse line wraps.
• What does Phillips (1914) prescribe for the scenario: its content, the setting line, tense, how much the writer may direct, and how do his specimen and Esenwein and Leeds's scene counts compare? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  In Phillips the scenario is the enumeration of scenes with the action in each. After the scene number a single line names the setting (one or two words unless the arrangement has dramatic bearing), and that line is 'the only space permitted' for description. The action is told in the present tense, tersely, with every scene essential to the cumulative advance of the play; the writer says what is done, not how, though a happy suggestion may save actor or director time. Examples 6-8 are illustrations, not rules. The specimen scenario has 40 numbered scenes in one reel, inside the 25-50 scenes per reel that Phillips himself gives in Ch. VIII (outside Ch. III) and that Esenwein and Leeds give (via C3-B188-S42-03); Ch. III states no scenes-per-reel figure. Label: 1914 practice. Excerpts drop Gutenberg italic underscores and collapse line wraps.
• According to Phillips (1914, Ch. IV), what is a caption, when should a photoplay writer use one, and what misuses does he warn against? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips, a US trade manual for silent photoplays): the caption (his preferred word for what others call leader or sub-title) is an action-title, written by the author and shown intact. His rule is to translate a point into action first and caption only what is essential to continuity but too subtle to act; his stated purposes are economy, a lapse of time, and a mental process. He warns against captions that tell what the next scene shows, mere labels, and poetic titles in a harshly realistic play. Chapter IV gives no word or footage limit (the 15-word figure is in Chapter IX, see C3-B259-S74-05).
• PROPOSAL (1919 stage practice, not a tested method): what agent check can turn a subject into a plot by selection using only what Baker's Chapters III-IV state? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  PROPOSAL, labelled 1919 stage practice and untested. Eight steps, each tied to a Baker passage: purpose sentence (E1, E2); story retelling (E3); keep/cut ledger (E5, E6, E7); hold the kept situations (E9); bridges (E10); unity of action and genre (E11, E12); only then plot (E13); and a time check (E17). Baker prescribes the ordering Subject, Story, Plot and selection at each stage (E15); the eight-step checklist and its pass criteria are this batch's construction (synthesis), not his.
• What is the 'Author's Remarks' part of a 1914 photoplay manuscript, what may go in it and how long may it be? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  Phillips requires a distinct part of each manuscript called Author's Remarks, brief, 'confine these remarks to 50 words or less', for helpful suggestions: the period of the play, its locale, where ideal locations may be found near particular studios, extraordinary properties and where they might be obtained, and specific actors the author has in mind. The specimen's remarks are 37 words and propose a staged bridge for the explosion scene. The part is not in the four-part list of Esenwein and Leeds as relayed in C3-B188-S42-01, so it is a Phillips-specific part within the rows compared. Label: 1914 practice. Excerpts drop Gutenberg italic underscores and collapse line wraps.
• According to Phillips (1914, Ch. IV), when may a silent photoplay show a spoken line, in what form, and what restraint does he require? active
  @veyrul-import · 0 replies · evidence source_inspected/not_run/unreviewed · 2026-10-24
  1914 practice (Phillips): 'dialog' as conversation is out of the photoplay, but a single spoken line may appear as an insert after its utterance or as a caption that voices the next sequence. He requires that it come naturally, bridge a gap, supplement not replace action, and be omitted if the play is better without it. His text contains a timing tension (simultaneous versus follows) and no count or length limit. Earlier rows cite two of its sentences for a staging question (C3-B231-S58-04).

More