Set image-to-image strength by what the output must preserve.

Name the source features you want to keep: silhouette, camera angle, color blocks, background layout, or facial arrangement. Image-to-image can preserve several kinds of information at once, so a vague goal makes evaluation difficult. A written preservation list gives you a basis for choosing strength.

Use a source you may process. Remove private information and avoid real-person sexual edits without permission. Lewdora supports fictional adult artwork, and the source must follow the same content and rights rules as a text-only request.

A light setting tends to preserve more of the source layout and detail. It suits controlled restyling or small visual changes, but it can resist a prompt that asks for a new camera angle or body arrangement. A balanced setting gives both signals room and often provides a useful first comparison point.

A strong setting can rebuild pose, identity, texture, and background. It helps when the source provides only a rough composition, but it can remove the feature you meant to preserve. Model behavior differs, so use the labels as test points rather than universal percentages.

Run the same source, prompt, model, dimensions, and other controls at three strength levels. Compare the result against the preservation list. Note where the source stops controlling the output and where the new prompt begins to appear with enough force.

Choose the lowest strength that delivers the required change. This rule preserves more source information and reduces unnecessary drift. If none of the three samples works, revise the source or prompt before testing smaller increments.

A source portrait and a prompt for a wide full-body scene ask for different framing. Higher strength may change the crop, but it may also discard the face and palette. Choose a source closer to the intended composition or separate the task into stages instead of forcing one request to solve both.

Keep style instructions coherent with the source. A sharp cel-shaded source and a soft painted prompt can produce mixed edges at balanced strength. Decide whether you want restyling, scene reconstruction, or detail repair, then write the prompt for that job.

Use character reference when recurring appearance matters more than the full source frame. Use pose reference when body structure matters more than texture or background. Use text-only generation when the source contributes no element you need to preserve.

These controls can work together on compatible models, but each extra signal creates another possible conflict. Add one control at a time and record its effect. The shortest successful workflow costs fewer tokens and remains easier to repeat.

Check identity, anatomy, text artifacts, background seams, and unintended source details. A strong transformation can invent content outside the source. A light transformation can retain marks, metadata, or identifying features you intended to remove.

Download approved work you need to retain. Lewdora’s gallery lives in your browser, so local cleanup can remove its records. Keep the source and generation settings with the output when you need to reproduce or explain the transformation.

Before choosing strength, list the source features the output must retain: broad composition, camera angle, silhouette, palette, facial identity, clothing shape, or background layout. Strength is not a universal quality control. It changes how much influence the source has relative to the prompt, and each model may define the direction or scale differently.

Use the product label and model documentation rather than assuming a number means the same thing everywhere. Save the source, prompt, model, dimensions, and value with every trial. If the model or provider changes, rebuild the sweep because a previously useful value may no longer represent the same balance.

Test a low, middle, and high source-influence setting with all other inputs fixed. Select the range that best preserves the required structure, then test one or two values inside that range. This is more efficient than adjusting by tiny increments from the start because the first pass identifies which part of the scale matters.

Compare the outputs against written criteria. Mark source layout, prompt adherence, identity, anatomy, and unwanted artifacts separately. A result that looks polished may still fail if it ignores the requested change. A rougher result may reveal the correct balance and only need a cleaner source or simpler prompt.

Higher source influence can preserve defects along with useful structure. Cropped hands, compression blocks, ambiguous limbs, watermarks, and noisy backgrounds may become harder to remove as the source carries more weight. Clean the permitted source, expand the canvas when the intended crop needs room, or draw a simple structural guide before increasing influence.

Do not use image-to-image as a way to bypass rights or consent. Transforming a protected image does not automatically make the result yours, and transforming a real person into sexual material without permission is prohibited. Use sources you created, licensed, or are otherwise allowed to process and publish.

A composition-preserving edit and a style change ask different questions. For composition, inspect subject placement, camera, silhouette, and negative space. For finish, inspect palette, linework, shading, and texture. The same strength may not serve both goals, so keep separate presets and notes.

When the prompt requests a large structural change, expect a different useful range from a light restyle. If no value preserves the required source feature while making the requested change, revise the source or split the job into stages instead of treating more attempts as the only option.