Quick answer
Give each reference one clear job. Use the approved character image for identity, a pose example for body position, and a style example for rendering. State which features to borrow and which to ignore. Keep the current panel’s cast separate from the story’s full character list.
For creators combining multiple visual references in an image-editing workflow. Reference roles here are planning instructions; they are not automatic model weights or guaranteed identity locks.
What should each reference control?
| Situation | Recommended approach | What to check |
|---|---|---|
| Character / identity | Face, hair silhouette and proportions | Ignore its old pose unless requested. |
| Outfit | Visible garment construction and material colors | Ignore a different wearer’s face. |
| Pose | Body position, gesture and weight distribution | Ignore identity, wardrobe and rendering. |
| Style | Linework, shading and paper or surface treatment | Ignore its people, objects and composition. |
Start with an approved source and one current action
Choose the image or panel that you intend to edit. Then write the scene in one sentence: who is present, what the main person is doing, and where the action occurs. This separates a current request from earlier narrative details.
For a solo shot, say that the approved protagonist is alone in the room. A companion may belong in the story but not this image. Avoid treating a cast bible as a list of everyone the generator must render. Keep historical scene notes out of the current action brief when they contradict it.
Assign roles instead of blending every detail
An identity image can define a face, hair silhouette and body proportions. A clothing image can define a jacket’s construction. A pose reference can define limb direction and weight distribution without lending its face or costume. A rendering reference can define pencil strokes or shading without adding its subject.
Write both borrow and ignore instructions. For example: “Use image 2 only for the seated body pose; ignore that model’s face, hair, clothing and background.” This does not create a technical control channel. It makes the intended division explicit for you and the model.
Resolve contradictions before generating
Two identity references may show different ages, hair lengths or costumes. Decide whether they are alternate designs or intentional story changes. Select the approved version for this panel and remove the rejected design. More references can add conflicting evidence instead of improving continuity.
A night-lit costume reference may obscure its base color. A pose image may show a hand hidden behind a prop. Describe the approved material color and required visible grip explicitly. Do not assume that a model can recover a hidden detail that you have never specified.
Keep reference numbering aligned with upload order
The organizer labels images in the order you add them and lets you change their roles. Copy the brief and upload the originals in that same order in the generation workspace. If you remove or reorder an image, regenerate the brief before use.
The generation model determines attachment limits. The organizer accepts up to three references to match the current shared generator’s visible upload capacity. A local role label does not upload an image, grant a paid attachment entitlement or bypass the model’s reference rules.
Reuse approved references, not every attempt
Keep one accepted identity reference and separate accepted costume versions. Store short descriptions of their intended roles beside them. When a generated result becomes the new approved source, record that decision so an older rejected image does not silently reenter a later request.
A useful reference set stays small enough to review. Compare the resulting character’s face, clothing and silhouette against the approved source, then check the action separately. Good identity with the wrong action still fails the task.
Step-by-step workflow
Write the current scene
List only the people needed now and one main action.
Add up to three clear references
Use the organizer to preview the images locally. Assign a role to each image.
Specify borrowed details
Describe which visible traits matter and which details should be ignored.
Resolve the conflict checklist
Choose one approved identity and costume. Remove contradictory references or rewrite the scene.
Copy and align
Copy the numbered brief. Upload the actual originals in that same order in the generator, respecting its plan and attachment limits.
Review by role
Compare identity, costume, pose and rendering separately before accepting a candidate.
A worked editing brief
Illustrative writing example. This is not a recorded before-and-after image test or a customer result.
A courier has an approved face and blue jacket. A separate pose image shows a different adult seated at a café, while the source panel uses loose pencil lines.
An ambiguous instruction
Combine these pictures and make the character consistent.
A more specific instruction
What would make the result acceptable?
Check the courier identity against image 1, seated weight against image 2, and stroke character against image 3. Reject an added café cup even if the face and pose are correct.
Review the result before accepting it
- The current scene names only the intended visible cast.
- Every reference has one explicit role and clear exclusions.
- Reference numbers match the actual upload order.
- Conflicting identities and costume versions are resolved.
- The generator’s attachment limits and plan prompts are respected.
If the edit still misses the mark
- The pose reference supplies the wrong face
- Restrict it to body position and repeat the approved identity anchors from the identity image.
- A person from the style reference appears
- Specify rendering traits and exclude that image’s subjects. Consider a subject-free style crop.
- The tool refuses another upload
- Use fewer references or follow the existing upgrade flow. Local organization does not change upload eligibility.
Common questions
Are reference roles actual model settings?
No. The organizer creates a clear human-readable instruction. It does not apply hidden weights, pose skeletons or identity locks.
Can I use more than three images?
This organizer supports up to three to match the shared generator’s visible upload capacity. Other model or project workflows may differ.
Are organizer previews uploaded?
The organizer reads previews locally and does not send its files to a server. Uploading them later in a generator is a separate action.
Should style and character always be separate images?
Not necessarily. One approved source may already contain both. Add another image only when it provides a missing, compatible constraint.
Continue your comic workflow
Remove an extra person from a comic panel, preserve the intended cast, and check the repaired background before replacing the original.
How to Keep Clothing Consistent Across Comic Panels →Build a useful costume reference, diagnose dark or changing clothing, and repair one garment without redesigning the character.
Redraw Comic Poses Without Losing the Character →Change a comic panel’s action, prevent outdated scene instructions from conflicting, and review the intended cast, costume and framing.