Face Photo - AI Front-Facing Conversion · öffentlicher Open Skill
Face Photo - AI Front-Facing Conversion
Face Photo - AI Front-Facing Conversion ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.
Other
5 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/u5cxen3y.jpg
# Face Photo - AI Front-Facing Conversion
Turns an uploaded photo whose face is turned sideways or at an angle into a natural, front-facing face photo (vertical 3:4), keeping the original person's features as intact as possible. No structured project state: inputs, prompts and results live in the conversation and as canvas image assets.
## When to use
- The user uploads (or points at) a photo and asks for a front-facing / head-on / "looking at the camera" face photo.
- The user asks to convert a profile or three-quarter view of a face into a frontal portrait.
- The user asks to redo, sharpen or slightly adjust a previously produced conversion.
## Input and output
- Input: one reference photo supplied by the user (attachment asset_id, preset asset `key`, `s3://` uri or image URL). If several images are present, ask which single photo to convert; do not merge multiple people into one face.
- Output: one image asset (portrait, vertical 3:4) placed on the canvas.
## Steps
1. Identify the single reference photo to convert and confirm it actually contains a face; if there is no face, say so and stop.
2. Before generating, load the visual conventions with `read_skill_file(name='face-front-facing-conversion', path='references/style-and-prompt.md')`. Do not write the prompt from memory.
3. Optional but useful when the face is heavily turned or the photo is noisy: call `recognize_image(image_url=<reference image>, prompt='Describe the person: face shape, hairstyle and hair color, apparent age, expression and mouth shape, skin tone, lighting, accessories.')` and carry those observations into the prompt as features to keep.
4. Compose a single English prompt from the template in `references/style-and-prompt.md`: front-facing conversion + strict identity preservation + the unified "face photo" look + vertical 3:4 head-and-shoulders framing. Never invent features that are not in the reference or in the app description.
5. Generate with the image-to-image channel:
`generate_image(prompt=<composed prompt>, input_urls=[<reference image>], model='i2i.hd.lite', orientation='portrait', aspect_ratio='3:4', resolution='1080P', name='Front-facing face photo', entity_type='face_photos')`
- `model` is always the channel `i2i.hd.lite` (reference image present, detail-preserving); never pass a raw model name.
- `aspect_ratio='3:4'` is used here because this app's output ratio is fixed at vertical 3:4; keep `orientation='portrait'` alongside it.
- Pass `resolution='1080P'` by default for a clean, high-detail face photo; if the platform returns `param_adjustments`, accept the adjusted value and continue.
- Always pass a short `name` and `entity_type='face_photos'`; never pass `group_id` / `index`.
6. If the result is clearly off — different person, over-retouched, or the head is still turned — retry once with the same input and a strengthened preservation clause from the reference file ("keep identity exactly", "no beautification", "no age change"). Do not loop more than once; if it still fails, tell the user honestly.
7. Reply in one or two sentences: what was converted and what stayed the same. Also state that parts the camera never showed are AI-inferred, so the result is not an exact reproduction of the real face. Do not paste long URLs; the asset is already on the canvas.
8. When the user is likely to want a next step (another photo, a sharper version, a different framing), call `set_guidance` in the same turn as the final text reply, and call no other tool in that turn.
## Notes
- Images only. If the user asks for a video from this photo, say this skill produces a still face photo and do not call `generate_video`.
- Keep the output a plain photographic face photo: no added props, jewelry, backgrounds, text/watermarks or style transfer that the app description does not mention.
- Refuse sexual, violent or otherwise policy-violating requests politely and do not call the generation tool.
- If `generate_image` returns `success=