Skip to content
Video of the day
Newsletter
Note—001DevelopingModels

Midjourney tests 'Thinking Mode': V8.2 re-checks its own image against your prompt, then regenerates

Midjourney is testing a Rerun (Thinking) button that makes V8.2 check a finished image against the prompt and generate again. It is Alpha-only, with no published benchmark, GPU cost or release date.

Drafted by The Generated · Filed by The Generated

Midjourney is testing a second pass for its image model. On alpha.midjourney.com, a finished V8.2 job now carries a "Rerun (Thinking)" button: the model looks at the image it made, finds where it missed your prompt, and generates again.

The misses it looks for are wrong objects, broken layout, bad anatomy and garbled text, which are also the usual reasons a storyboard frame gets thrown out. Midjourney says its tests show gains in prompt accuracy, typography and coherence, though it has published no numbers, and the rerun works on both standard and edit jobs made with V8.2.

Two things are still unknown: how much GPU time a thinking rerun costs, and when it leaves the Alpha site. Midjourney says it may later ship it as a general thinking mode or as a way to add more thinking after a job.

In practice, if you build keyframes or boards where the layout and the sign on the wall have to be right, try the rerun on a frame that almost worked before you rewrite the prompt, and send Midjourney the cases where it fails.

Source: Midjourney announcement and Alpha changelog, Oct 8, 2026.

Provenance — Midjourney: Testing Thinking Mode; Midjourney Alpha changelog 10/7/26; Midjourney on X · Status: single-source, reported
An image model that checks its own work against your prompt, then tries again
Note—002DevelopingModels

A developer used Claude to clone Adobe's suite as free open source

FilmCraft, a free open-source editor modeled on Premiere and built with Claude, rates itself only 50 to 60 percent ready for real work in its own roadmap, with H.264 the only delivery export and no plugin support.

Drafted by The Generated · Filed by The Generated

ArtCraft, a studio that sells AI image and video tools, has put seven free, open-source apps on GitHub that copy the look, panels and shortcuts of Adobe's suite. FilmCraft is the one built to feel like Premiere Pro, and CEO Brandon Thomas says he wrote the whole suite in Rust with Claude Opus 5.5.

The most useful document is FilmCraft's own roadmap, updated Oct 5. It puts the feature checklist at about 87 percent and rates it ready for real work at about 50 to 60 percent. H.264 is the only delivery export, plugin support is effectively missing, and AAF and OMF handoffs to Avid and Pro Tools have never been validated. The roadmap also admits that some of its parity scores were written by the agents doing the work.

Clean room is Thomas's description rather than an established fact, and Adobe had not commented as of Oct 8. The genuinely new part is that FilmCraft exposes more than 650 commands over MCP, an open protocol that lets AI assistants operate software, so an agent can cut, grade and export without anyone touching the mouse.

In practice, it costs nothing to try: cut a rough assembly, export H.264 and round-trip an FCPXML file into Resolve or Premiere before you trust it with a client job.

Source: FilmCraft README and roadmap on GitHub, Oct 8, 2026.

Provenance — Ars Technica — "Software is over": Bold AI developer takes aim at Adobe with open source clones; PetaPixel — Someone rebuilt free, open-source versions of Photoshop, Premiere and Lightroom using AI; Phoronix — AI makes "open-source, clean room" implementations of Photoshop & Premiere in Rust; AfterD · Status: verified: HIGH (Ars Technica, PetaPixel, Phoronix, AfterDawn — independent)
The free Premiere clone built with Claude rates itself half ready
Note—003DevelopingModels

Claude learns motion design: Anthropic ships Claude Motion, and Runway, Luma and Higgsfield plug in on day one

Claude Motion builds editable animated explainers from a prompt without a video model, and seven video tools including Runway and Luma open the result. It is a beta on Team and Enterprise plans only, with no published pricing or usage limits.

Drafted by The Generated · Filed by The Generated

Anthropic launched Claude Motion in beta on Oct 8, and it moves a chunk of motion graphics work into the chat window. You ask for a 30-second explainer, an animated chart or a product walkthrough, and Claude writes code that animates your own text, charts, shapes and images. Every word, number and timing stays editable, and the result downloads as an MP4.

Anthropic is explicit that no video generation model is involved, so there is no generated footage and no AI-generated people. That gap is where the video tools stepped in on day one. Motion pieces open in Runway, Luma, Adobe, Descript, HeyGen, Higgsfield and invideo, with Canva and Captions listed as coming soon. Runway pitches itself as the next step, adding generated footage, voiceover and music and recutting a 16:9 piece for vertical, while Luma offers to resize, refine and version it.

The limits matter here. Motion is in beta on Claude Team and Enterprise plans only, Enterprise admins have to switch it on, and Anthropic has not published pricing or usage limits.

In practice, a client on a Team plan can now draft its own explainer before calling anyone. Our read, and it is analysis: the paid motion work shifts toward the finish, meaning generated footage, sound, pacing and the versions for every channel.

Source: Anthropic, Runway and Luma announcements, Oct 8, 2026.

Provenance — Anthropic: Build live dashboards and animate explainers with Claude; The Verge: Claude can now turn data into animated explainers; Reuters via US News: Anthropic launches dashboard, animation tools for Claude; Runway: Runway is in Claude Motion; Runway on X; Luma on X; Higgsfield car commercial demo · Status: verified: HIGH (Anthropic + The Verge + Reuters; Runway and Luma named launch partners)
Motion graphics move into the chat window, and video tools become the finishing step

That’s all of Models.

9 Oct · 3 notes, filed as the record moved.

Models — archive

Also filed

Hollywood & Film1 note
Grimes' team says AI was one tool among 70+ artists
Grimes' 'Accelerando' video used AI 'sparingly', her team says, with 70+ artists on the crew

Grimes' team says AI was a supplementary tool in the Accelerando video, made by more than 70 artists over two to three months. Which shots used AI, and which tools, has not been disclosed.

Creators & AI Talent1 note
A found-footage home video from 2003, made with Seedance 2.5
Meem: Seedance 2.5 on Higgsfield AI

Meem's 30-second found-footage clip, made with Seedance 2.5 on Higgsfield AI, could easily pass for a real early-2000s home video. The only giveaways are a few odd prompted actions, like moving a peg on an empty washing line.

Brands & Advertising2 notes
Disney bids $20M for Character.AI's code, and hires the people who wrote it
Disney bids $20M to be the stalking horse for Character.AI's assets as the chatbot startup files for Chapter 11

Disney bid $20 million to license some of bankrupt Character.AI's code, while most of the startup's staff joined Disney. The app, its users and their chats are not part of the deal, and rival bids can still top it.

Follow The GeneratedX (opens in a new tab)Instagram (opens in a new tab)TikTok (opens in a new tab)YouTube (opens in a new tab)