P Image Edit

Prerequisites

Install and load these skills before generating (skip if already in context via @pruna):

Or install the full suite once: npx skills add PrunaAI/pruna-skills@pruna -y

Follow each skill's Before generating / craft sections — do not restate guide content here.

Agent habit

In the first reply, name `p-image-edit` in backticks, confirm PRUNA_API_KEY, then ask for the source photo (and edit brief). Open intake → generation-diversity clarification intake when needed. Draft edits with Prompt craft (dynamic + faithful). Redirect to p-image-try-on when the job is garment fit from packshots.

Prompt craft (dynamic + faithful)

Edit prompts are surgical and request-locked. Change only what the user asked; keep everything else identical. Diversity applies to how you phrase the change — not to inventing a new scene.

Fidelity check (before pay): the prompt must still satisfy the user’s change and every stated keep. If a keep is missing, add it. Do not “spice” the background/cast when they asked for a background-only swap.

When NOT to use

Use a different skill instead:

HTTP (curl)

Upload references

curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: ${PRUNA_API_KEY}" \
  -F "content=@/path/to/reference.png"

Use each response urls.get in input.images.

Create (async — recommended)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-image-edit' \
  -d '{
    "input": {
      "prompt": "Change background to soft gradient, keep subject identical",
      "images": ["https://api.pruna.ai/v1/files/FILE_ID"],
      "aspect_ratio": "9:16"
    }
  }'

Poll and download: follow pruna-api.

Complete the random seed ritual from generation-diversity before writing prompts — do not pass the ritual string as API seed. Optional seed only when the user requests reproducibility.

Create (sync — quick test only)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-image-edit' \
  -H 'Try-Sync: true' \
  -d '{
    "input": {
      "prompt": "Change background to soft gradient, keep subject identical",
      "images": ["https://api.pruna.ai/v1/files/FILE_ID"],
      "aspect_ratio": "9:16"
    }
  }'

Edit flow

Follow generation-diversity still-image prompt flow (edit section):

1. Lock change + keeps — user’s surgical ask; formula in Prompt craft (dynamic + faithful).
2. Upload refs — POST /v1/files; urls.get → input.images (1–5). Never invent file URLs.
3. Ritual seed → draft edit prompt (image-prompting edit craft) → fidelity check (every keep clause present).
4. Confirm — prompt, refs, aspect_ratio, turbo (off for hard edits).
5. POST — async curl below; checklist before upscale/video.

Aspect ratio: prefer match_input_image for edits; output should keep the plate aspect — if drift occurs, retry with the same ref URL and explicit keep clauses.

Pruna note: avatar pipelines — edit from locked upscaled hero URL; upscale again before p-video* — never pass raw edit URLs to video models.

Not edit: new subject from scratch → p-image. Garment fit from packshots → p-image-try-on.

Multi-reference (up to 5): index refs by role — e.g. three-reference composite: face from image 1, outfit from image 2, cafe from image 3; turbo: false for hard composites. Full pattern in image-prompting p-image-edit-prompting.

Required input

• prompt (string)
• images (array of 1–5 URLs, typically https://api.pruna.ai/v1/files/{id})

Common optional fields

• aspect_ratio: match_input_image, 1:1, 16:9, 9:16, etc.
• turbo (boolean, default true; turn off for harder edits)
• seed, disable_safety_checker

Typical next steps

Common follow-ons after this skill: