Fix hands and faces

Repair common AI portrait failures

Use this skill when you need a repeatable image workflow, not a one-off prompt guess. It defines the input, output, tool path, and quality gate before generation starts.

Input

broken image

Output

repaired image

Use this skill Copy options

Best fit

Repair failures

ADetailerImpact PackGPT Image 2

High-intent repair page

Job to be done

Create repaired image from broken image, using a task-first workflow instead of a loose prompt dump.

Workflow type

inpainting, detailer

Quality promise

The output must preserve user intent, survive visual inspection, and be ready for the target channel.

Workflow outline

Prepare the source input: broken image.
Pick a first tool path: ADetailer, Impact Pack, GPT Image 2.
Generate or edit toward the promised deliverable: repaired image.
Review the image for drift, broken details, unreadable text, and weak composition.
Save the prompt, references, and final settings that produced the best reusable result.

Copy-ready skill card

Skill: Fix hands and faces

Scene: Fix & Edit / Repair failures

Input: broken image

Output: repaired image

Tools: ADetailer, Impact Pack, GPT Image 2

Example input: Product image: generated bottle visual with warped label edge

Issue to fix: distorted label text and uneven cap shape

Must preserve: composition, lighting, product color, material

Edit scope: repair only the broken areas

Do not change: claims, logo, flavor, size, or packaging structure

Quality check: the result must match the promised output and preserve the user intent.

Branch logic

If the user has not provided broken image, ask for it before generating the final repaired image.
If a reference image is provided, preserve the subject identity, proportions, and important details before changing style or scene.
If the first result fails visual inspection, fix the smallest broken area first instead of regenerating the whole image.
If the first tool path fails, switch to the next listed option: Impact Pack, GPT Image 2.
If the task depends on masks, text, or object edges, inspect those details at full size before accepting the output.
If the image will be published commercially or publicly, run the risk guardrails before export.

When not to use

Do not use this as proof that the generated repaired image is factually real or legally approved.
Do not use it when the user needs professional legal, medical, safety, or regulated-compliance judgment.
Do not publish the result until rights, consent, authenticity, and platform constraints have been checked.

Copy distribution

Take this skill into your AI tool as a prompt, a workflow card, or a portable SKILL.md draft.

Quality gate

The final asset clearly matches the requested repaired image.
The main subject still reflects the original broken image.
Text, hands, logos, edges, and product details are not visibly broken.
The result is usable in the target channel without another full regeneration.
Confirm the user has rights or consent to use the source photo and avoid exposing private personal details.