Skip to content
Note—001DevelopingModels

ByteDance's Seedream 5.5 promises 100 rounds of image edits that leave the rest of the frame alone

ByteDance's new image model claims stable edits across 100 rounds and transparent output, and Higgsfield added it the same day. Every score is ByteDance's own, and there is no ByteDance price outside China yet.

Drafted by The Generated · Filed by The Generated

Most AI image edits carry a hidden cost: ask for one change and the model quietly redraws everything around it. ByteDance has built its new model around fixing that, which matters to anyone who spends more time repairing frames than making them.

Seedream 5.5 launched on Monday in Doubao and Dreamina in China, with developer access due this week. ByteDance says it can keep editing one image across 100 rounds while untouched areas stay the same down to the pixel, with no color drift. It also outputs real transparent backgrounds, full and partial, so titles, props and cutouts drop into a compositing timeline as ready layers instead of being masked by hand.

Higgsfield added the model the same day and says US access is coming soon. ByteDance has published no price outside China, where one launch example puts a 2K edit at about 0.28 yuan, roughly 4 US cents.

The evidence so far is all ByteDance's. Its 100-round demo shows an image surviving a long chain of changes, not one character held steady: the man at the chessboard has become someone else by edit 59, and by edit 99 the frame is a chess piece. Its own chart still rates most film and TV frames unusable, and puts Google's Nano Banana 2.1 ahead on realism when editing an existing photo.

Our take: edit-first is the right bet for filmmaking, because the pain is rarely the first draft and usually the twentieth note. Run one character still through 30 rounds before rebuilding a storyboard around it.

Source: ByteDance Seed, Nov 9, 2026; Higgsfield on X; Sina Tech.

Read the full article: thegenerated.com/stories/bytedance-seedream-5-5-100-round-editing-20261109

Provenance — Vendor blog: https://seed.bytedance.com/en/blog/more-realistic-images-more-stable-editing-across-100-rounds-introducing-seedream-5-5; Model page: https://seed.bytedance.com/en/seedream5_5; Sina Tech: https://finance.sina.com.cn/tech/it/2026-11-09/doc-iniycrzu2908981.shtml; Jiemian: https://m.jiemian · Status: verified 10 Nov 2026: built from the published article
ByteDance's Seedream 5.5 is built to survive 100 rounds of edits
Models

Adobe will let your own AI agent edit Premiere and Photoshop projects, but not until 2027

Adobe previewed local connectors that let agents such as Claude or ChatGPT edit native Photoshop, Illustrator, Premiere and After Effects files, keeping layers and sequences editable. Nothing ships until a private beta in early 2027, and real Premiere timeline edits are so far only a stage demo.

Drafted by The Generated · Filed by The Generated
Adobe MAX 2026 key art: a stylized portrait of a figure wearing a chrome orchid headpiece against a red background

Adobe on Tuesday 10 November previewed local connectors that let outside AI agents such as Claude or ChatGPT edit native Photoshop, Illustrator, Premiere and After Effects projects on a user's own computer. None of it ships yet. Adobe says a private beta opens in early 2027, and it has published no price, agent list or tool list.

The shift here is plain: the editable project file, not the flattened render, is becoming the thing AI works on, and Adobe now accepts that the agent a creator already uses can be the front door, with Photoshop and Premiere as tools it calls. What most coverage blurred is that outside agents reaching Adobe tools is not new. Adobe already ships cloud connectors for Claude, ChatGPT and Gemini, and an unofficial open source bridge built by an Adobe community staffer has let Claude drive Photoshop since March 2025. The new part is official, local control of your real files.

Keep reading — 5 min

That’s all of Models today.

1 note, filed as the record moved.

Models — archive

Also filed

Cinematographers push back on greybox shoots
Cinematographers at Camerimage push back on greybox shoots, where AI adds the lighting after the shoot

At Camerimage, cinematographers say greybox shoots, where actors are filmed on plain sets and AI adds the sets and lighting later, move the look of a film into post. The Bitcoin detail rests on one IndieWire line.

Microdrama studios say the hook, not the shoot, keeps a vertical series alive
Microdrama studios say the first 30 seconds decide a vertical series, not the shoot

Bump and İki Dakika told TCCF that hooks and villains hold a vertical feed, not production budgets. Both shoot live action, every quote comes from one session report, and Bump's scale figures are its own.

CAA and Ogilvy will make celebrity ads from licensed digital doubles
CAA and Ogilvy launch ECHO, letting brands make AI ads from licensed celebrity scans

Ogilvy's ECHO builds ads from scans kept in The CAA Vault, and the star must approve every asset before release. Terry Crews is the only reported talent so far, no brands are named, and the cost savings are the companies' own estimates.