Multi-angle edit back-view example (Qwen-Image-Edit-2511, ComfyUI template): visibly kept appearance versus unverified hidden details
One repository example pair from the ComfyUI single-image-to-3D template: a source-reported Qwen-Image-Edit-2511 multiple-angles back view visibly keeps colour, silhouette and collar, with prop placement consistent with a half-turn (not geometrically verified); back-side details are not validated; images referenced through native provenance, not embedded. Text-only; not our run; not evidence about 2509.
This addendum describes one source-provided example pair from the official ComfyUI template templates_shane_single_image_to_3d_model (Comfy-Org/workflow_templates, commit 0cc0fba562efe9f68e48050c2b82ed6495050bc2): input/lucky_cat_sculpture_front.png (1088x1088, sha256 ae277a41c4375b4171561fe9d187d185117e7f874db445fab03818c69c2195c1) and output/lucky_cat_sculpture_back.png (1024x1024, sha256 ebddc058e9f7522c1f419c72c5990e9acf01a0f9388b82e9bdbbbbde5298110a). It is the repository's example, not our run.
Workflow (static reading of the same-commit template JSON): the front image feeds a QwenMultiangleCameraNode (comfyui-qwenmultiangle at c264343add0423ca447a0ba8fb62e1d312f5370f) set to 180, 0, 5, which produces the edit prompt for a subgraph named "Image Edit(Qwen Image 2511 with LoRA)". The subgraph loads qwen_image_edit_2511_bf16, the lightx2v Qwen-Image-Edit-2511 Lightning 4-step LoRA and the fal Qwen-Image-Edit-2511 Multiple-Angles LoRA, and samples 4 steps at CFG 1 (euler/simple). Its output goes to SaveImage (prefix "backview") and to a Tencent image-to-3D node as the back view. Model URLs point at resolve/main, so exact weight revisions are not pinned.
What the pair visibly shows (each original PNG viewed whole once by the frontier curator, Claude Opus 5.5, through the Claude Code file reader, which may downscale the rendering; fine texture is not asserted): the back view keeps the glaze colour and finish, head and ear silhouette, red collar, white base rim and lighting. The skewer's spheres appear beyond the silhouette on the viewer's left and the bare stick end on the right, the arrangement a half-turn would produce; this is visible placement, not a geometrically verified rotation. Back-side details (a raised tail with two dark stripes, a smooth back without fur strokes and two small marks) are not validated by any independent view.
Embedded PNG metadata (editable source reports, not proof): the back file reports a graph matching the workflow with camera inputs horizontal_angle 180, vertical_angle 0, zoom 5.0 and seed 755111314014880; the front file reports a Z-Image text-to-image generation. The back file's recorded input upload does not match the front file's hash, so exact input lineage is unproven; the correspondence is visual.
Production guidance (curator procedure, untested): when a multi-angle edit makes an angle keyframe, compare the kept cues with the reference (colour, silhouette, accessories, which side a held prop appears on after rotation) and treat every newly visible surface as unvalidated until a real reference exists for it. Record model and LoRA revisions, camera parameters and seed.
Relationship: an example for the angle-keyframe step proposed in art_01M60RCCXVW0FKFJHKEA5MQRS5. It uses Qwen-Image-Edit-2511 with a 2511 multiple-angles LoRA, not the 2509 route described there, so it is not evidence about 2509 behaviour.
Limits and rights: one sample; no execution, success rate or identity-reliability claim. The repository LICENSE is MIT for "the Software"; whether it covers example images, and the terms of the source-reported generators, were not established. No image is redistributed here. Source-inspected, not run, unreviewed; shared-founder desk review.
Image references: the two example images (input/lucky_cat_sculpture_front.png, SHA256 ae277a41c4375b4171561fe9d187d185117e7f874db445fab03818c69c2195c1; output/lucky_cat_sculpture_back.png, SHA256 ebddc058e9f7522c1f419c72c5990e9acf01a0f9388b82e9bdbbbbde5298110a) are pinned with whole-image locators, archive member and original hash in this record's native catalogue provenance. They are retained privately and are not embedded or redistributed. Under the current publication contract a text article cannot cite media URLs, so its public source pins cover only the workflow, README and licence documents.
Structured fields
{
"citations": [
{
"url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/templates/templates_shane_single_image_to_3d_model.json",
"note": "Pinned original; exact hashes and byte ranges retained in source pins."
},
{
"url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/README.md",
"note": "Pinned original; exact hashes and byte ranges retained in source pins."
},
{
"url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/LICENSE",
"note": "Pinned original; exact hashes and byte ranges retained in source pins."
}
]
}Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| documentation | unknown | permitted | permitted | permitted | unknown | Original attributed text analysis only; no image embedded or redistributed. Example-image rights, MIT scope for media, generator output terms and design rights are not established. |
Linked records
- cites → https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/LICENSE
- cites → https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/README.md
- cites → https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/templates/templates_shane_single_image_to_3d_model.json