On 8 October Anthropic put a motion-graphics tool inside Claude. Claude Motion writes the animation as code, so every word, number and timing stays editable, and it exports an MP4 or hands the file to Luma, Higgsfield or Runway to finish. No video model, no generated people. That is the lead because text, charts and packs in motion are what fill most brand social calendars, and until Thursday no chat tool shipped that format.
Claude Motion: the animation is code, and Luma picks it up
Claude Motion is in beta on Team and Enterprise plans, off by default for Enterprise until an admin enables it. It animates the text, charts, shapes and images you give it, lets you change any word, number or timing in the editor, and downloads as an MP4. Anthropic says it uses no video-generation model, so there is no generated footage and no AI-made people; handoff partners are Adobe, Descript, HeyGen, Higgsfield, invideo, Luma AI and Runway (Anthropic, 8 Oct). Luma’s connector opens the animation in Luma, keeps the motion while restyling the look, and reframes it to 9:16, 1:1, 4:3 or 21:9 (Luma, 8 Oct). The same day Docs, Slides and Design left beta on every plan; the standalone Design site closes on 14 December.
My read: I built One Group. Every Shelf. exactly this way: 50 seconds, eight categories, one grid, one tile movement, the motion written in code rather than in a timeline app. Code is why eight sections read as one system. Change the timing once and every tile follows. That is what this hands a brand team with no motion designer. Two limits. It animates what you give it, so the pack you upload is the pack you get; feed it approved artwork, never a prompt for the pack. And it is Team and Enterprise only, so a freelancer on Pro gets the handoff partners, not the tool.
Higgsfield Katana cuts a video from one photo, inside Claude
Higgsfield’s changelog for 8 October lists Katana, a Claude skill that builds a video edit from one reference photo in Aura Farming, Motion or Clipping styles. You can use up to five reference images, pick presets from three categories and steer with a prompt. Users can publish community presets, a 3 to 30 second motion video plus up to 40 images, and shared presets go live immediately. It runs from the Katana page or by asking Claude through the Higgsfield MCP (Higgsfield changelog, 8 Oct).
My read: Aura farming is a TikTok format: a slow push-in, a hard cut on the beat, a character held like a hero. From one photo, it is the cheapest mascot content a brand can make. The rule I hold is the one from Festival Toffee, where the packs stood in real 3D and the model never redrew one: use Katana for the character, never for the pack. Two more. Every edit is metered in credits; I cap a first test at 50 credits before it goes near a brief. And a community preset is public the second you publish it, so an unreleased pack never goes into one.
Midjourney tests a Thinking Mode that checks its own image
Midjourney is testing a Thinking Mode on its alpha site: rerun a V8.2 job, standard or edit, and the model checks the image for missed prompt details such as wrong objects, layout, anatomy or text, then regenerates. Midjourney says the goal is prompt accuracy, typography and coherence, calls it an early version, and gives no speed or cost; it is enabled from the lightbox with “Rerun (Thinking)” (Midjourney, 8 Oct). The same release adds shared folders with Collaborator and Viewer roles and link sharing (Midjourney alpha changelog, 8 Oct).
My read: Typography and anatomy are the two reasons I finish packaging and hands outside the model. A second pass that reads its own output against the prompt is the right idea; it is what a retoucher does. But it is a rerun, so budget two generations where you budgeted one, and test it on your own failure cases, not on a demo set. The shared folders matter more than they sound. A brand folder where the agency generates and the client only views is the approval loop most teams fake over WhatsApp.
What I’d do today
If your team has a Claude Team plan, take the chart from last week’s brand review and have Motion animate it, then send it to Luma for a 9:16 cut and watch what breaks. On Pro, the news is the handoff partners, Luma, Runway and Higgsfield, because that is where the file lands. Need the production side? Here is how I work.
Sources
- Anthropic (claude.com) — Build live dashboards and animate explainers with Claude: Claude Motion beta, partners, Design site closing 14 Dec, 8 Oct 2026 — claude.com
- Luma AI — Start in Claude Motion. Finish it in Luma: connector, restyle, reframe 9:16/1:1/4:3/21:9, 8 Oct 2026 — lumalabs.ai
- Higgsfield — Creator Hub changelog: Katana release, presets and community presets, 8 Oct 2026 — higgsfield.ai
- Midjourney — Testing Thinking Mode, 8 Oct 2026 — updates.midjourney.com
- Midjourney — Alpha changelog 10/7/26: thinking mode on V8.2, shared folders, 8 Oct 2026 — updates.midjourney.com