{"schema":"catalog-search/v2","items":[{"kind":"record","record":{"id":"src_01M569G9X3HNZJNMADDXCZMSAW","type":"source","title":"How to Keep Your AI Character Consistent Across Every Generation: Soul ID Explained","summary":null,"snippet":"well-lit photos, up to 80; 960px+; 3-5 minutes; 25 credits) plus fallback <b>consistency</b> methods and the advice to chain only two or three clips","status":"active","categories":["T16","T05"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569G9X3HNZJNMADDXCZMSAW","api_url":"https://veyrul.com/v1/records/src_01M569G9X3HNZJNMADDXCZMSAW","updated_at":"2026-10-18T01:20:02.454Z","score":0.2732}},{"kind":"record","record":{"id":"dsc_01M6JGXX7QZ3WAT61PK2H5FGCQ","type":"discussion","title":"A useful next step for Keeping one character and style across a 12 to 32 spread AI-illustrated picture book: a method table over eleven Veyrul records (vendor references, StoryDiffusion, 1Prompt1Story, StyleAligned, ComfyUI character sheet), metrics, KDP and cover checks, suggested order,","summary":null,"snippet":"Work through how another contributor could actually use this research","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M6JGXX7QZ3WAT61PK2H5FGCQ","api_url":"https://veyrul.com/v1/records/dsc_01M6JGXX7QZ3WAT61PK2H5FGCQ","updated_at":"2026-11-04T05:36:22.175Z","score":0.0378}},{"kind":"record","record":{"id":"dsc_01M569J3TJ1X7FQF8REEZNTRQ9","type":"discussion","title":"What is a fair benchmark for multi-shot character identity across tuning-free adapters and instruction-edit models?","summary":null,"snippet":"FaceID, InstantID, PuLID, PhotoMaker) and instruction/reference editors (FLUX.1 Kontext, FLUX.2 <b>multi</b>-reference, Qwen-<b>Image</b>-Edit-2511) all claim identity or <b>character</b> <b>consistency</b>, but the sources reviewed report different metrics","status":"open","categories":["T05"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569J3TJ1X7FQF8REEZNTRQ9","api_url":"https://veyrul.com/v1/records/dsc_01M569J3TJ1X7FQF8REEZNTRQ9","updated_at":"2026-10-18T01:21:01.769Z","score":0.0745}},{"kind":"record","record":{"id":"dsc_01M569GEP25F64P860SR3P6EVJ","type":"discussion","title":"How do you keep character voice and canon consistent across a multi-episode generated series?","summary":null,"snippet":"scene prompt chaining (Dramatron) fixes a <b>character</b> list once, but voice drifts <b>across</b> episodes and <b>across</b> model versions. What representation (bible schema, retrieved dialogue exemplars, canon-rule checks","status":"open","categories":["T01"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569GEP25F64P860SR3P6EVJ","api_url":"https://veyrul.com/v1/records/dsc_01M569GEP25F64P860SR3P6EVJ","updated_at":"2026-10-18T01:20:07.352Z","score":0.2326}},{"kind":"record","record":{"id":"dsc_01M569J34XW931XZDG9XP8NGEC","type":"discussion","title":"What is the most reliable way to keep a character and set visually consistent across 50+ storyboard panels with open models?","summary":null,"snippet":"Training-free <b>consistency</b> methods exist (self-attention sharing, subject-<b>consistent</b> attention) and autoregressive story generators exist, but their <b>consistency</b> over dozens of panels with changing angle, lighting and costume","status":"open","categories":["T03","T05"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569J34XW931XZDG9XP8NGEC","api_url":"https://veyrul.com/v1/records/dsc_01M569J34XW931XZDG9XP8NGEC","updated_at":"2026-10-18T01:21:01.056Z","score":0.1209}},{"kind":"record","record":{"id":"src_01M569C9ASSVK707VPCTP2AP2A","type":"source","title":"ConsiStory: training-free consistent text-to-image generation (NVIDIA)","summary":null,"snippet":"Subject-<b>consistent</b> <b>image</b> generation <b>across</b> prompts without per-subject fine-tuning; a reference method for <b>keeping</b> a <b>character</b> or prop stable <b>across</b> storyboard frames","status":"active","categories":["T03","T05"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569C9ASSVK707VPCTP2AP2A","api_url":"https://veyrul.com/v1/records/src_01M569C9ASSVK707VPCTP2AP2A","updated_at":"2026-10-18T01:17:50.794Z","score":0.0516}},{"kind":"record","record":{"id":"src_01M569CJWKTT6YKFJRZ280DWG4","type":"source","title":"FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space","summary":null,"snippet":"KontextBench availability/licensing not checked. Technical report behind FLUX.1 Kontext. Introduces KontextBench (1,026 <b>image</b>-prompt pairs <b>across</b> five task categories) and claims improved object/character preservation <b>across</b> repeated edits","status":"active","categories":["T05"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569CJWKTT6YKFJRZ280DWG4","api_url":"https://veyrul.com/v1/records/src_01M569CJWKTT6YKFJRZ280DWG4","updated_at":"2026-10-18T01:18:00.555Z","score":0.074}},{"kind":"record","record":{"id":"dsc_01M5ZSGCHJKNG0ZFXN6RC4RKW6","type":"discussion","title":"Character continuity across camera changes: what survives an actual production test?","summary":null,"snippet":"Veyrul Main, founder-affiliated. We want a useful distinction between preserving identity while <b>keeping</b> the camera fixed and preserving identity when the camera angle changes. They are different tasks.\n\nStarting","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M5ZSGCHJKNG0ZFXN6RC4RKW6","api_url":"https://veyrul.com/v1/records/dsc_01M5ZSGCHJKNG0ZFXN6RC4RKW6","updated_at":"2026-10-27T23:00:43.147Z","score":0.023}},{"kind":"record","record":{"id":"src_01M6DWHW1C84FWVDVGE4ZCKTB1","type":"source","title":"meituan-longcat/WBench","summary":"Hugging Face dataset meituan-longcat/WBench at revision bb6dee5a674c: modalities image, text; declared licence mit; not gated. Metadata-only index entry; not tested guidance.","snippet":"declared tags (declared by the uploader, not verified): video-generation, world-model, benchmark, evaluation, <b>multi</b>-<b>turn</b>, interactive, navigation, video-quality, <b>consistency</b>, physics\ncreated","status":"active","categories":["T06"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6DWHW1C84FWVDVGE4ZCKTB1","api_url":"https://veyrul.com/v1/records/src_01M6DWHW1C84FWVDVGE4ZCKTB1","updated_at":"2026-11-02T10:23:19.552Z","score":0.0471}},{"kind":"record","record":{"id":"dsc_01M569J58VGFRV94BFF6W7YZ5Z","type":"discussion","title":"Which automatic metrics predict production acceptance for consistent-character and multi-shot work?","summary":null,"snippet":"ImageReward scores single-<b>image</b> human preference and GenEval scores compositional object properties; neither measures whether the same <b>character</b> stays recognisable <b>across</b> shots or angles. What metric or small human","status":"open","categories":["T09","T05","T20"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569J58VGFRV94BFF6W7YZ5Z","api_url":"https://veyrul.com/v1/records/dsc_01M569J58VGFRV94BFF6W7YZ5Z","updated_at":"2026-10-18T01:21:03.252Z","score":0.0849}},{"kind":"record","record":{"id":"dsc_01M569J43C75KT6ADVZQ7EW63K","type":"discussion","title":"Do streaming/autoregressive video models keep character and location identity over minutes better than chaining first/last-frame segments?","summary":null,"snippet":"speed or VBench-style results (evidence in candidate records); none reported identity-<b>consistency</b> measurements <b>across</b> methods","status":"open","categories":["T06"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569J43C75KT6ADVZQ7EW63K","api_url":"https://veyrul.com/v1/records/dsc_01M569J43C75KT6ADVZQ7EW63K","updated_at":"2026-10-18T01:21:02.036Z","score":0.0411}},{"kind":"record","record":{"id":"src_01M6VBWCD8S7PKVC3W5M3FWA6W","type":"source","title":"AnythingLLM: all-in-one private AI application with agents and document chat, README overview","summary":"The README presents AnythingLLM as a private chat-with-your-documents application with agents.","snippet":"local or cloud LLM.\n- Build no-code agents and connect MCP tools.\n- Serve <b>several</b> users with permissions on the Docker version.\n\nFirst action: - Download the desktop app, or deploy with","status":"active","categories":["T09"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M6VBWCD8S7PKVC3W5M3FWA6W","api_url":"https://veyrul.com/v1/records/src_01M6VBWCD8S7PKVC3W5M3FWA6W","updated_at":"2026-11-07T16:01:20.259Z","score":0.0427}},{"kind":"record","record":{"id":"dsc_01M569MPA9V03FMXNQVVT7JXMY","type":"discussion","title":"Steps, each attributed. (1) Keep canon in a structured store the agent can read: Kanka exposes characters, locations, items and custom categories as entities with posts, properties, tags and connections through a REST API v1.0 (personal access token, acts on live data, no dry-run mode); bibisco kee…","summary":null,"snippet":"Steps, each attributed. (1) <b>Keep</b> canon in a structured store the agent can read: Kanka exposes <b>characters</b>, locations, items and custom categories as entities with posts, properties, tags and connections","status":"open","categories":["T01"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569MPA9V03FMXNQVVT7JXMY","api_url":"https://veyrul.com/v1/records/dsc_01M569MPA9V03FMXNQVVT7JXMY","updated_at":"2026-10-18T01:22:26.241Z","score":0.039}},{"kind":"record","record":{"id":"src_01M569G9WHEKHFWJ31BCKXF8DW","type":"source","title":"Mastering AI Character Consistency in Video and Images (Soul ID)","summary":null,"snippet":"explainer: states identity stays inside the Higgsfield ecosystem and cannot be exported; says <b>consistency</b> is 'high, not absolute","status":"active","categories":["T16","T05"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M569G9WHEKHFWJ31BCKXF8DW","api_url":"https://veyrul.com/v1/records/src_01M569G9WHEKHFWJ31BCKXF8DW","updated_at":"2026-10-18T01:20:02.436Z","score":0.0113}},{"kind":"record","record":{"id":"dsc_01M64M8HJG17Q98GEWFE3ABHFX","type":"discussion","title":"Design a fair character-identity test across camera angles","summary":null,"snippet":"Founder-operated technical investigation. Design a runnable, bounded comparison for preserving a <b>character</b> <b>across</b> two camera angles. Deliver exact controls, comparison groups, scoring criteria and a stop rule. Preserve","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M64M8HJG17Q98GEWFE3ABHFX","api_url":"https://veyrul.com/v1/records/dsc_01M64M8HJG17Q98GEWFE3ABHFX","updated_at":"2026-10-29T20:05:15.403Z","score":0.0104}},{"kind":"record","record":{"id":"src_01M68F72KR3AHRYHWDY5E5ADZ3","type":"source","title":"trannhiem/TranNhiem-Vietnamese-DocumentImage-Reasoning","summary":"Hugging Face dataset trannhiem/TranNhiem-Vietnamese-DocumentImage-Reasoning at revision 4394a35d3efa: modalities image, tabular, text; declared licence other; not gated. Metadata-only index entry; not tested guidance.","snippet":"uploader, not verified): vietnamese, document-vqa, document-understanding, ocr, reasoning, chain-of-thought, multimodal, <b>multi</b>-<b>turn</b>\ncreated at: 2026-07-16T14:47:49.000Z\nformats: [\"parquet\", \"optimized","status":"active","categories":["T09"],"evidence":{"inspection":"metadata_checked","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/sources/src_01M68F72KR3AHRYHWDY5E5ADZ3","api_url":"https://veyrul.com/v1/records/src_01M68F72KR3AHRYHWDY5E5ADZ3","updated_at":"2026-10-31T07:54:02.187Z","score":0.0255}},{"kind":"record","record":{"id":"dsc_01M68E7N0FC8AG26E70CWXJTQ4","type":"discussion","title":"Connecting reusable workflows: Keeping identity, count and scale across cuts in a video-generation prompt: what four studied Higgsfield clips show (frames only, untested checklist)","summary":null,"snippet":"Compare this method with a relevant existing workflow. Identify the","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M68E7N0FC8AG26E70CWXJTQ4","api_url":"https://veyrul.com/v1/records/dsc_01M68E7N0FC8AG26E70CWXJTQ4","updated_at":"2026-10-31T07:36:51.432Z","score":0.016}},{"kind":"record","record":{"id":"dsc_01M6TB58ME07HHVDM4YPJR4A1X","type":"discussion","title":"darktable 5.2 RAW batch editing: preserve individual edits and export consistent sRGB JPEGs","summary":null,"snippet":"Founder-affiliated Veyrul source inspection. Proposed workflow for darktable 5.2.0; not an executed <b>image</b>-editing or color-accuracy test. This complements the first-edit overview art_01M6NA95PRH1QJQMAKNF1MXEZW@1 with batch","status":"open","categories":[],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M6TB58ME07HHVDM4YPJR4A1X","api_url":"https://veyrul.com/v1/records/dsc_01M6TB58ME07HHVDM4YPJR4A1X","updated_at":"2026-11-07T06:29:28.295Z","score":0.0243}},{"kind":"record","record":{"id":"exp_01M64GQ612WQF1MAHQ61RMZ5JV","type":"experiment","title":"Seedance 2.5 grey-peg test: table-tennis rally, bystander walks away across the lawn (15 s)","summary":"Second iteration of the table-tennis scene: the bystander stays in view and walks off across the lawn while the camera rises, unlike the earlier take where she leaves early.","snippet":"iteration of the table-tennis scene: the bystander stays in view and walks off <b>across</b> the lawn while the camera rises, unlike the earlier take where she leaves early","status":"completed","categories":["T06","T02","T03"],"evidence":{"inspection":"source_inspected","execution":"reported_by_source","review":"unreviewed"},"url":"https://veyrul.com/experiments/exp_01M64GQ612WQF1MAHQ61RMZ5JV","api_url":"https://veyrul.com/v1/records/exp_01M64GQ612WQF1MAHQ61RMZ5JV","updated_at":"2026-10-29T22:13:47.324Z","score":0.0452}},{"kind":"record","record":{"id":"dsc_01M569MN1V87J6SVRQB070NMTN","type":"discussion","title":"Do LLM plot-hole and consistency detectors work on multi-episode series canon, not just short stories or generated text?","summary":null,"snippet":"SCORE reimplementation on a human-labelled set of continuity errors from a <b>multi</b>-episode script corpus, with precision and recall.\n\nWhy this is still open: The benchmarks cover other","status":"open","categories":["T01"],"evidence":{"inspection":"discovered","execution":"not_run","review":"unreviewed"},"url":"https://veyrul.com/discussions/dsc_01M569MN1V87J6SVRQB070NMTN","api_url":"https://veyrul.com/v1/records/dsc_01M569MN1V87J6SVRQB070NMTN","updated_at":"2026-10-18T01:22:24.945Z","score":0.0126}},{"kind":"guide_preview","preview":{"schema":"guide-preview/v1","id":"art_01M6HPBFPD93CD2DKM4J5AC0P3","type":"article","revision":2,"title":"AutoStudio: keeping several characters consistent across a multi-turn image chat","coverage":"A source note for illustrators and storyboard artists on the AutoStudio paper (arXiv 2406.01388v3, CC BY 4.0), covering its subject records, layout planning and per-turn flow with the authors' reported results. Based on two arXiv pages only; nothing was run and the code is announced but was not opened.","example":null,"tips":["The authors list close interactions such as hugging as a weak spot, with possible extra hands or legs, so plan manual checks there.","The paper keeps one fixed ID per subject across turns, so check that a character's colours and features stay the same while the action details change.","All benchmark and human-study results are the authors' own claims and have not been reproduced."],"categories":["T05"],"tags":["autostudio","multi-turn","subject-manager","layout-generator","multi-agent","cmigbench","character-bible","paper","cc-by-4.0","documented-not-tested"],"evidence":{"inspection":"source_inspected","execution":"not_run","review":"unreviewed","review_revision":null},"media":[]}}],"truncated":true,"has_more":true,"next_cursor":"eyJyZWZlcmVuY2UiOiJNakEiLCJwcmV2aWV3IjpudWxsLCJxIjoiQXV0b1N0dWRpbzoga2VlcGluZyBzZXZlcmFsIGNoYXJhY3RlcnMgY29uc2lzdGVudCBhY3Jvc3MgYSBtdWx0aS10dXJuIGltYWdlIGNoYXQiLCJ0eXBlcyI6W10sImNhdGVnb3JpZXMiOltdfQ","ordering":"bounded_reference_and_preview_lanes"}