Which AI features upload my image?

There's one test: does the feature hand that layer to a large cloud model? The ones that do (these 10: Old Photo Restore, AI Colorize, Replace background, ID Photo, Upscale, Beautify, Decorate and De-age, plus Split into editable layers and Edit text in image) send that layer (as a JPEG, scaled down first if it's large) to the cloud, and the result comes back to the canvas (most replace the layer; Split into editable layers builds new layers and keeps the original hidden; Edit text in image first duplicates the original layer as a hidden backup, then builds text layers and fills the old text on the original layer with background) — the image is saved to the uploads folder of this account's TutuDrive (when TutuDrive is available, this site saves it first; otherwise it goes out with the message and TutuAgent saves it; if you're not signed in, it's a temporary visitor account) and then handed to the assistant; the ones that don't (Remove BG, Content Fill, Brighten & color, Smart upscale 2×, Expand Image, Content-Aware Scale (whole image), Compress to size) all run in your own browser — even Remove BG (AI cutout) is an in-browser model. “Generate” and “Generate & export PSD” sit in between: they run in the cloud, but creating the picture sends only the sentence you wrote (when the canvas already has content, the assistant takes a thumbnail of it to see what's there). Exactly which ones is listed in the table above on this page, and it's generated from code — nothing gets left out.

Which ones run entirely on my machine and never send the image out?

The whole local group works, with no sign-in and no upload: Remove BG (AI cutout), Content Fill, Brighten & color, Smart upscale 2×, Expand Image, Content-Aware Scale (whole image) and Compress to size. What sends anything out are the two cloud groups: text only (Generate, Generate & export PSD; when the canvas already has content, the assistant takes a thumbnail of it to see what's there) and picture-sending (these 10: Old Photo Restore, AI Colorize, Replace background, ID Photo, Upscale, Beautify, Decorate and De-age, plus Split into editable layers and Edit text in image). None of them needs an API key, and no sign-in first — there are free credits every day. Separately, while you're signed out your work stays on this device; once you sign in, work in the editor is saved automatically to your own cloud projects — that's saving, not AI processing.

Does the AI cutout need a network connection?

No. It is an in-browser model: the model file downloads once on first use, and after that it runs on your machine, and nothing is uploaded during the cutout. That is also why it keeps working offline — provided the model has been downloaded before.

Do any of these cost money?

The editor itself costs nothing. The local group is entirely free, unlimited, watermark-free and works offline (AI cutout needs a connection the first time to download a model of about 100MB). The two cloud groups need no sign-up and have free credits every day (no key to bind), metered centrally by TutuAgent; when they run out, the assistant tells you directly and shows you how to keep going. Print orders are paid per order — that ships a physical photo to your door, which is buying a thing, not buying software.

Can the AI run on my own computer?

Yes. Install TutuNode and the AI assistant can hand work to your computer — free, no usage counting — taking over automatically on cloud refusals, queues or used-up credits.

AI features · what runs locally and what goes to the cloud

Three groups, stated plainly: local / cloud, text only / cloud, image sent. None needs sign-up; the two cloud groups have free credits every day.

or start a blank document

"Has AI" tells you nothing. The question worth asking is which parts run on your machine and which parts send your picture away.

So this page does just one thing — sort every AI feature by those two questions. Group one, computed locally: AI cutout uses an in-browser model (the model file downloads once on first use, then it works offline too), while Content Fill, Brighten & color, smart upscale, outpaint, content-aware scale and compress-to-size are plain pixel algorithms. This group is the part that still works with the network cable pulled — unlimited and watermark-free. Group two, cloud, text only (no sign-in, like group three, with free credits every day): AI text-to-image and batch generation send only the sentence you wrote to create the picture, and a new image comes back; when the canvas already has content, the assistant takes a thumbnail screenshot to look at it. Group three, sends the picture: these send the layer you selected (as a JPEG; large images are downscaled first) to the AI as an attachment, and the result comes back to the canvas: most replace that layer, while “Split into editable layers” builds new layers and keeps the original hidden, and “Edit text in image” first duplicates the original layer as a hidden backup, then builds text layers and fills the old text on the original layer with background. Ctrl+Z undoes any of them. Exactly which ones is listed in the table below — generated from code, so a new feature shows up there automatically and you'll never get "not mentioned in the copy, but it uploads anyway". For the full chat assistant (file mentions, tool cards, model choice), go to the assistant page — no sign-in there either.

Which ones send your image out

Computed in your browser · the image never leaves

Unlimited, no watermark; works offline too (Remove BG needs a connection the first time to download a model of about 100MB). Not one byte of your image leaves while it's processed.

Remove BG
Content Fill
Brighten & color
Smart upscale 2×
Expand Image
Content-Aware Scale (whole image)
Compress to size

Computed in the cloud · text only

Creating the picture sends only the sentence you wrote; when the canvas already has content, the assistant takes a thumbnail screenshot to look at it.

Generate
Generate & export PSD

Computed in the cloud · sends your image

Sends the layer you picked to the cloud. Most replace that layer; “Split into editable layers” adds new layers and keeps the original hidden; “Edit text in image” first duplicates the original layer as a hidden backup, then adds text layers and fills the old text on the original layer with background. All undoable.

Old Photo Restore
AI Colorize
Replace background
ID Photo
Upscale
Beautify
Decorate
De-age
Split into editable layers
Edit text in image
“✨ Change this to…” on a selection (AI local edit — sends the area you circled)
Sketch to Photo (sends the sketch)

These three groups are generated straight from the code (the where and attach fields of the action table) — not a hand-written list. A new feature shows up in its group automatically, so this page cannot end up claiming less than what actually happens.

Does it cost anything? How credits work, item by item

Key features

The local tier is the biggest

Cutout, content-aware fill and scale, upscale, expand, sky replace — all in the browser.

Only three send your image

No "and so on", no round numbers — that table is generated from the code, so a new feature shows up by itself.

Text-to-image sends only the sentence

Creating the picture sends only the sentence you wrote, and a new image comes back; when the canvas already has content, the assistant takes a thumbnail screenshot to look at it.

Cloud ones send your image

Anything that calls a large model is computed in the cloud, with no sign-up and free credits every day. The 7 local features are unaffected: no sign-in, no limit, and they work with the network off.

How to use

1
See whether this tier suffices

Cutout, fill, upscale, and expand are all local — no sign-in, no upload.
2
Use text-to-image to generate

Write the prompt under “Generate” in the AI panel; creating the picture sends text only. When the canvas already has content, the assistant takes a thumbnail screenshot to look at it.
3
Old photo repair needs no sign-in either

Restore, colourise and background swap run in the cloud — no sign-up, free credits every day, and no key to bind.
4
Results land as layers

Generated images land as a new layer by default; restore, colorize and the like replace the layer you selected, and Ctrl+Z undoes it. Either way it's on the canvas and you can keep editing — not a flat picture you can only download.

FAQ

Which AI features upload my image?: There's one test: does the feature hand that layer to a large cloud model? The ones that do (these 10: Old Photo Restore, AI Colorize, Replace background, ID Photo, Upscale, Beautify, Decorate and De-age, plus Split into editable layers and Edit text in image) send that layer (as a JPEG, scaled down first if it's large) to the cloud, and the result comes back to the canvas (most replace the layer; Split into editable layers builds new layers and keeps the original hidden; Edit text in image first duplicates the original layer as a hidden backup, then builds text layers and fills the old text on the original layer with background) — the image is saved to the uploads folder of this account's TutuDrive (when TutuDrive is available, this site saves it first; otherwise it goes out with the message and TutuAgent saves it; if you're not signed in, it's a temporary visitor account) and then handed to the assistant; the ones that don't (Remove BG, Content Fill, Brighten & color, Smart upscale 2×, Expand Image, Content-Aware Scale (whole image), Compress to size) all run in your own browser — even Remove BG (AI cutout) is an in-browser model. “Generate” and “Generate & export PSD” sit in between: they run in the cloud, but creating the picture sends only the sentence you wrote (when the canvas already has content, the assistant takes a thumbnail of it to see what's there). Exactly which ones is listed in the table above on this page, and it's generated from code — nothing gets left out.
Which ones run entirely on my machine and never send the image out?: The whole local group works, with no sign-in and no upload: Remove BG (AI cutout), Content Fill, Brighten & color, Smart upscale 2×, Expand Image, Content-Aware Scale (whole image) and Compress to size. What sends anything out are the two cloud groups: text only (Generate, Generate & export PSD; when the canvas already has content, the assistant takes a thumbnail of it to see what's there) and picture-sending (these 10: Old Photo Restore, AI Colorize, Replace background, ID Photo, Upscale, Beautify, Decorate and De-age, plus Split into editable layers and Edit text in image). None of them needs an API key, and no sign-in first — there are free credits every day. Separately, while you're signed out your work stays on this device; once you sign in, work in the editor is saved automatically to your own cloud projects — that's saving, not AI processing.
Does the AI cutout need a network connection?: No. It is an in-browser model: the model file downloads once on first use, and after that it runs on your machine, and nothing is uploaded during the cutout. That is also why it keeps working offline — provided the model has been downloaded before.
Do any of these cost money?: The editor itself costs nothing. The local group is entirely free, unlimited, watermark-free and works offline (AI cutout needs a connection the first time to download a model of about 100MB). The two cloud groups need no sign-up and have free credits every day (no key to bind), metered centrally by TutuAgent; when they run out, the assistant tells you directly and shows you how to keep going. Print orders are paid per order — that ships a physical photo to your door, which is buying a thing, not buying software.
Can the AI run on my own computer?: Yes. Install TutuNode and the AI assistant can hand work to your computer — free, no usage counting — taking over automatically on cloud refusals, queues or used-up credits.

Try AI features · what runs locally and what goes to the cloud now

No install · no sign-up · local stays local · free AI credits every day