Upscale image

The Upscale image mode is the correct workflow when you want a genuinely larger final image while keeping scene, subject, and framing as stable as possible.

This is the right workflow when you want to:

• bring a small file to a more useful size for e-commerce, presentations, or larger outputs;
• recover perceived detail while increasing final size;
• avoid using a normal edit workflow as a pseudo-upscaler.

In this mode the main control is the upscale factor (from 1x to 4x). The prompt is optional.

Use Upscale image if:

• you want a bigger photo, not just a cleaner one;
• the source file is too small for the final use case;
• you want to increase detail and readability without changing scene and composition;
• you want a larger deliverable while keeping the original frame.

• If you need to change specific parts of the image: use Single image edit.
• If you only want better quality, cleanup, or sharpness without truly increasing final size: use Enhance photo quality.
• If the photo is old, damaged, or visibly deteriorated: use Restore photo.
• If you want a larger image but also a more creative reinterpretation: use Generate image from text or Single image edit with an explicit prompt.

• 1 image.
• An upscale factor from 1x to 4x.
• (Optional) A prompt to guide detail recovery or rendering.
• (Optional) Extra controls depending on the selected model.

When processing starts, the app:

• uses the uploaded image as the source;
• applies the selected upscale factor;
• sends any optional prompt and model-specific controls;
• calculates pricing from the estimated output megapixels, not from 1K/2K/4K tiers.

Expected behavior:

• the composition tries to remain coherent with the source;
• the model adds perceived detail while increasing final size;
• if the source image is much smaller than the requested output, part of the detail may be reconstructed, so the result can become more interpretive than strictly faithful.

If the uploaded file exceeds the app input limit, it is reduced before sending.

In this mode you typically see:

• Mode: workflow selector (choose Upscale image).
• Image upload: upload the photo.
• Upscale model: choose the upscaler available in your installation.
• Upscale factor: from 1x to 4x.
• Estimated output: preview of the expected final megapixels.
• Optional prompt: use only when you want to steer detail recovery.
• Model-specific controls: may include creativity, resemblance, CFG, inference steps, variants, face enhancement, and detail controls.
• Upscale: start processing.
• Use as input: reuse the result for later refinement.

1. Select Upscale image.
2. Upload the cleanest, least compressed file you have.
3. Start with 2x if you are unsure.
4. Leave the prompt empty unless you need to steer a specific behavior.
5. Touch advanced controls only if the selected model truly needs them.
6. Generate the result.
7. Check the file at 100% zoom and judge whether the recovered detail looks believable.

Practical strategy:

• if in doubt, do a more conservative first pass;
• a believable 2x upscale is often better than an over-reconstructed 4x result;
• if the result is good but has local defects, refine those in Single image edit.

Usually you can leave it empty. Use it only when you want to guide:

• skin texture;
• fabrics and materials;
• detail style;
• cleanup level or realism.

Examples:

• “Keep skin texture natural, avoid plastic skin.”
• “Recover fabric weave realistically, without artificial sharpening.”
• “Clean studio look, believable detail, no artifacts.”

Tip: keep it short and specific. In upscaling, the prompt should guide detail recovery, not rewrite the whole scene.

Some models expose extra sliders or options. In general:

• higher Creativity = more reconstructed detail, but also more reinterpretation risk;
• higher Resemblance = stronger fidelity to the source;
• higher Guidance / CFG = stronger prompt adherence;
• higher Inference steps = more refinement work, but also more time;
• Face enhancement = improves faces when supported by the model;
• Model variants = change the recovery behavior, from more conservative to more generative.

Not every control is always visible: it depends on the selected model.

• This is the correct workflow when you truly need to enlarge an image.
• If the input is much smaller than the requested output, the model may reinterpret parts of the photo: for example, a full-body subject can become a tighter or more portrait-like framing.
• Higher factors increase together:
  • final size,
  • credit cost,
  • processing time,
  • risk of reconstructed detail.
• Upscale pricing follows output megapixels, so two images with the same factor may cost differently if they start from different source sizes.

• Lower the upscale factor.
• Remove the prompt or make it more conservative.
• If available, choose a more conservative variant or raise resemblance.

• If the model exposes Face enhancement, lower it or turn it off.
• Avoid aggressive prompts.
• Start from a lower factor.

• Reduce creativity or detail controls, if available.
• Ask explicitly for “natural texture” in the prompt.
• Check whether a lower factor gives a more believable result.

• Increase gradually from 2x to 3x or 4x.
• Remember that an already large source image may show a subtler change at the same factor.

• Remember that pricing depends on the estimated output megapixels.
• The larger the source image and the higher the factor, the higher the final cost.

• If you need local corrections, switch to Single image edit.
• If you want a cleaner render without changing composition, you can do a light refinement in Enhance photo quality.
• If the image is still too weak, do not jump straight to the maximum factor: use a more conservative strategy or start from a better source.