Choose the reference that controls the part you want to change.

Choose character reference when the same fictional adult needs to remain recognizable across new scenes. The source should show the face, hair silhouette, palette, and signature traits that matter. A clean portrait offers more identity detail than a distant figure inside a crowded frame.

The reference does not freeze the character. A new angle, expression, costume, or light can change perceived identity. Keep a stable text block for recurring traits and compare several outputs. Treat the reference as guidance rather than a copy command.

Choose pose reference when body position, limb arrangement, or composition defines the scene. A source with a readable silhouette gives the model a stronger structural cue. Heavy occlusion, loose clothing, extreme perspective, or cropped limbs can make the pose hard to interpret.

The pose source should not define the new character. Describe the fictional adult, clothing, environment, and visual finish in the prompt. Inspect hands, limb contact, joints, and camera perspective because generative systems can reinterpret the source structure.

Use character reference when the face, hair, or palette drifts. Use pose reference when the output ignores the stance or places limbs in the wrong region of the frame. Use image-to-image when you want to preserve more of a complete source composition while transforming its details.

Prompt edits still matter. A reference cannot resolve a text conflict such as asking for a seated pose while the pose source stands upright. Read the prompt and source together before spending another generation. Remove instructions that compete for the same visual decision.

A compatible model can receive character and pose guidance in one request. Use a character source that exposes identity and a pose source that exposes structure. Keep both images simple enough that each signal has one job. Similar camera angles can reduce the amount of reinterpretation the model must perform.

Start with one reference, confirm its effect, then add the second. If you enable both from the first run, you cannot tell which source caused a conflict. Record the selected model and control settings because support and behavior can differ across the live catalog.

Use sources you created, licensed, or received permission to process. A pose can still come from a photograph of a person, and a character reference can still expose a real identity. Source rights and consent apply even when the output uses an anime finish.

Lewdora prohibits content involving minors, age-ambiguous sexual subjects, non-consensual sexual material, and unpermitted sexual deepfakes. Build fictional adult references for repeat projects and keep a record of where each source came from.

Score identity with face shape, hair, eye color, palette, and signature details. Score pose with silhouette, limb placement, orientation, and framing. Separating the criteria prevents a strong pose match from hiding character drift or a recognizable face from hiding broken anatomy.

Keep failed outputs in the sample set when you evaluate a workflow. Publish only claims that the recorded inputs support. Lewdora’s methodology fixes prompts and settings, changes one variable, and names limitations so a selected success does not stand in for repeat performance.

Use a character reference when the face, hair, palette, or other identity anchors keep changing. Use a pose reference when the body arrangement, camera-facing direction, or silhouette fails. Use image-to-image when the source composition itself should remain recognizable while the prompt changes its content or finish. The controls overlap, but they do not ask the model to preserve the same information.

Name the failure before uploading anything. If the identity works and the body does not, adding a second identity image can strengthen the wrong signal. If the pose works and the character drifts, increasing structural guidance can lock the unwanted body without repairing the face. A specific diagnosis prevents unnecessary inputs and token spend.

Choose a source with a clear silhouette, visible joints, and limited overlap between limbs. A neutral background makes body structure easier to read. Extreme foreshortening, loose clothing, furniture, and cropped hands can hide the information the pose control needs. The source does not need the target character or finish; it needs legible structure.

Confirm you have permission to process the image. A pose reference does not grant permission to reproduce a real person or place that person in sexual material. For adult content, use your own drawings, licensed pose resources, or synthetic references that depict fictional adults. Keep source-rights notes with the project.

Start with the text-only baseline, then add the character reference, then add the pose reference if the selected model supports both. Compare each stage using the same prompt and dimensions. This sequence shows what each input changed and exposes conflicts between identity, clothing, pose, and camera instructions.

When the combined result degrades, remove one reference before changing prompt language. The pose image may imply a crop that hides the character anchors, or the character image may imply a head angle that conflicts with the pose. Simplify the sources, choose references closer to the desired camera, or split the shot into a less demanding composition.

Reference support belongs to the active model, not to the account in general. Check which controls appear after every model change and read the displayed token cost before submitting. A comparison that silently switches models cannot show whether the reference type or the model caused the difference.

Save the model, prompt, dimensions, source identifier, reference role, and result together. Do not store a sensitive source in a shared note. A short rights record stating who created the source, which license or permission applies, and where the original is kept is enough for many personal workflows.