LongCat Edit: change one locked still (15cr)

LongCat Edit is an edit. Point at the object. A new text-to-image throws away the still you approved.

Meituan, 15 credits, category edit-image, requires_image: true. Description: "Longcat Edit is an image editing model that supports long-context image understanding and precise editing based on detailed textual instructions. It excels at complex multi-step edits and nuanced image transformations." Max reference: 1. No listed max resolution, no aspect enum, no qualities. audio does not apply. 15 credits is a flat megapixel-priced call. You are buying a delta on a file that already survived review. You are not buying a second pack member that "looks similar."

Long context is for instructions, not a new scene

The catalog's claim is long-context understanding and precise editing from detailed textual instructions, including complex multi-step edits. That is permission to write a careful delta: change the jacket, then the signage, then the practical light — on the same still. It is not permission to paste a 200-word scene description that competes with the pixels.

Point at the object. Name what stays. The input image is ground truth. A new text-to-image generate will redraw the face, the crop, the type. That is cheaper on some stills rows and more expensive in review, because you then have to re-approve identity. LongCat is the 15-credit alternative to that redraw.

AI photo editor is the door. Edit a photo with AI is the agent. Best image editing model is the ranked shelf. There is no Meituan provider hub. This page is LongCat Edit only.

One locked still, one reference slot

requires_image is true. max_reference is 1. You attach the plate. You do not attach a moodboard of five products and hope for a composite. If the job is a composite, you are on a different edit SKU with a larger reference budget. If the job is "this approved hero, with three written changes," you are here.

15 credits is not a 2-credit generate. It is an edit price. Using it as a generator — uploading a random stock photo and prompting a full new campaign — is how you pay edit rates for a text-to-image job. Using a 2-credit generate to "fix" an approved hero is how you throw the hero away.

No resolution is listed. Do not promise 4K from this record. Inspect the output at 100%. Multi-step in one instruction is the listed strength; multi-step still means each step is a change to the same still, not a new still per step unless you choose to iterate.

Throw away the still only on purpose

Teams re-generate because the edit UI feels slower than a prompt box. The still they already approved had the right face, the right crop, the right legal line in the frame. A new t2i will not keep those for free. LongCat's job is to keep them on purpose.

If the still was never a keeper, do not edit it. Generate. If it was a keeper except for one (or a sequenced few) objects, edit. 15 credits. Point. Do not narrate a different photograph.

FAQ

Does LongCat Edit require an image?

Yes. requires_image is true. There is no silent fallback to text-to-image. No still, no job.

Why is this 15 credits?

That is the catalog figure for this edit row (flat, per-megapixel billing). You are paying for a long-context instruction edit on a locked file, not for a 2-credit sample of a new identity.

Can I attach several reference images?

The record lists max_reference: 1. One plate. Write the rest of the complexity into the instruction. If you need many image inputs, pick an edit model whose catalog lists a higher cap.

When should I generate a new still instead?

When identity is still open — wrong person, wrong layout, wrong idea. LongCat will not rescue a still you would not ship. It will spend 15 credits making a precise change to whatever you uploaded. Upload a keeper.