Changelog
What’s new in Beemm Vision
What changes in Beemm Vision, month by month: the models joining the catalogue, new canvas, Generate and MCP connector features, and the fixes you will notice. Each entry is dated the day it went live and, when there is one, links to the page that details it.
Updated
November 2026
- Model
Gemini Omni Flash 1.2: text, image, references and video editing
Google's Gemini Omni Flash 1.2 heads the Gemini Omni Flash line in Generate and on the canvas: text or start image (with an optional end image) to video, up to 10 reference images and 3 reference clips, and editing of a clip of up to 15 seconds. 360p to 4K, 3 to 15 seconds. Two companions: Extend adds 3 to 10 seconds to an existing video, and Upscale raises a clip of up to 15 seconds to 720p, 1080p or 4K.
- Model
Kling O3 Standard and Pro, Kling V3 Motion Control
Kling O3 now comes in Standard and Pro: text, image (start and end) or references to video, 3 to 15 seconds, with or without sound. The O3 video reference also comes in both tiers, and O3 video editing gains a Standard tier (Pro stays the default). New too, Kling V3 Motion Control: a character from an image takes on the movement of a video, up to 30 seconds. In Generate and on the canvas.
- Model
Lip-sync and relighting on the canvas
Eight new canvas models. Make a portrait speak over an audio track with Kling AI Avatar v2, MiniMax H3 Max Lip Sync, OmniHuman v2 or sync-3 Avatar; re-sync the lips of an existing video to a new voice with sync-3 Lipsync, Kling LipSync or HeyGen Lipsync; relight a video from a lighting reference with MiniMax H3 Max Relight. The cost follows the exact length of the audio or video, measured before anything is charged. Audio in WAV or M4A (WAV only for OmniHuman v2): convert an MP3 first.
- Model
Seedance 2.0 up to 4K, Seedance 2.5 drafts
Seedance 2.0 Standard now goes from text or from images (start and end) to video, from 480p up to 4K, 4 to 15 seconds. On the canvas, Seedance 2.5 Draft renders a 480p preview, and the Seedance 2.5 Complete node regenerates the same take in 1080p, within 14 days. Also new: text to video on Seedance 1.5 Pro, and Seedance 1.0 Pro Fast (480p to 1080p, 2 to 12 seconds).
- Model
Ideogram Vector: from text or an image to SVG
Ideogram Vector generates SVG files from a prompt, with up to 5 reference images, and converts an existing JPEG, PNG or WebP image into an SVG, detailed or simplified. Also new: Ideogram V4 Tiling, for seamless textures and patterns that repeat edge to edge. In Generate and on the canvas.
- Model
HeyGen Video, Mischief 1, Boreal and Pixelcut Looping Video
Four video models from four new editors, in Generate and on the canvas. HeyGen Video: text, image or references to video, with sound, up to 2K. Mischief 1 by The Biological Computing Co.: image to video, 3 or 5 seconds. Boreal by Creatify: text to video, with an optional image and audio track, up to 20 seconds. Pixelcut Looping Video: a product photo becomes a seamless loop of 5 to 15 seconds.
- Feature
Layerize: a Seedream 5 Flash engine
The Layerize node now offers Seedream 5 Flash next to Seedream 5 Pro, which stays the default: the same split into layers, cheaper, and at the same price whatever the resolution. The node shows the cost before you run it.
- Model
Kling 4.0 Flash, the economical Kling 4.0
Kling 4.0 Flash joins the Kling 4.0 line in Generate (the Flash version) and on the canvas: text, start image or references (up to 10 images and 5 clips) to video, plus video editing. 720p, 3 to 20 seconds, sound always included, no end image.
- Model
GPT Image 2.6, Flare and Sunburst
OpenAI's GPT Image 2.6 is available in its two variants, Flare and Sunburst, for generation and editing (up to 16 input images and a mask). Low, medium or high quality, 1K, 2K or 4K, eight ratios. In Generate and on the canvas.
- Feature
From the site to Generate, on the right model
The button on a model page, or on a guide about a model, now opens Generate with that model already selected, after sign-in if needed. A plan chosen on the pricing page is kept through sign-up and leads to its payment once the required steps are done.
- Fix
MCP connector: generation is restored after reconnecting
When you reconnect, a single consent screen gives the Claude or ChatGPT connector access to generation, edits and workflow runs again, instead of leaving it read-only. If your connector no longer generates, remove it in Réglages → Applications connectées (Settings → Connected apps), then reconnect it.
- Feature
New sign-in page for the CLI
Signing in from the beemmvision CLI opens a page that shows the account in use, what the CLI will be able to do and a "Ce n'est pas moi" (This isn't me) button, then commands ready to copy once the terminal is connected. You can check the account and permissions before authorizing your terminal.
- Fix
Steadier layout on Generate and on mobile
Generate tiles now take each image's real shape right away, and the mobile navigation bar no longer shifts the page when it opens. The page moves much less under your finger or cursor while it loads.
- MCP
Image edits from the MCP connector
The remote MCP connector now offers Beemm Vision's four image edits: upscale, background removal, inpaint and expand, with the result saved to your gallery. You can edit an image from a conversation with Claude or ChatGPT without opening the app.
- Fix
Finished generations are no longer lost after a server incident
If our server stops after the model provider has finished a generation started from Generate, the result is now recovered and delivered to your gallery instead of being marked as failed. You receive the image or video you asked for even when the incident is on our side.
- Fix
Clear Google sign-in errors
Errors from the Google sign-in button are now shown in plain language, such as a lost internet connection or an account that already exists with that address, instead of a technical message. You know right away what to do to sign in.
October 2026
- MCP
Image batches from the MCP connector
The remote MCP connector starts up to 12 images in one call, with, in claude.ai, a free card that sums up the batch and its cost. You can produce several variants of a visual in a single request.
- MCP
Run a workflow from the MCP connector
The remote MCP connector can now run a workflow (beemm_run_workflow tool), where it could previously only build one. Your assistant can assemble a workflow, run it and follow the result without you opening the app.
- Model
Genjutsu Light: motion transfer and object swap
Genjutsu Light Motion Transfer and Genjutsu Light Object Swap (Higgsfield) join Generate and the workflow as the Light version of Genjutsu, for source videos of up to 15 seconds. You can transfer a motion onto a character or swap an object in a short clip.
- Feature
New Editor node
The Editor node stacks up to 12 images, or every layer from a Layerize node through a single link, and renders the composition on Run, downloadable as PNG, ZIP, PSD or SVG. You can assemble a poster or product visual from several generations without leaving the workflow.
- Feature
Layerize: export layers as ZIP, PSD or SVG
Layers produced by Layerize can be downloaded as PNGs with their positions (.zip), as a Photoshop file (.psd) or as SVG for Illustrator and Figma. You can pick the composition up in your usual software without cutting the image out again.
- Model
Kling 4.0 now available
Kling 4.0 (text, image or references to video) and Kling 4.0 · Video Edit are open in Generate, in the workflow video node and in the MCP. You can generate or edit a clip with the new Kling version without switching tools.
- MCP
The MCP connector works with ChatGPT
ChatGPT can connect to the Beemm Vision MCP connector. You can read your projects, workflows and images from a ChatGPT conversation.
- Model
FLUX 3 Max and PixVerse C1 for video
Two video models arrive: FLUX 3 Max (text, image or keyframes to video) and PixVerse C1 (text, image, transition between two images or references to video). They widen the choice for animating an image or linking two shots.
- Model
Nano Banana 2.1, Krea 3, FLUX.3 Klein, Luma Ray 3.2 and more
Nano Banana 2.1, Krea 3 and FLUX.3 Klein (plus Klein Base) for images, and Vidu Q4, Kandinsky 6.0 Lite, Happy Horse 1.0 and 1.1, Luma Ray 3.2 and Bernini-R for video, join Generate, the workflow and the catalogue. You can try them on the same prompt as your usual models and compare the results.
- Feature
Credit top-ups for subscribers
Subscribers can buy credit packs on top of their plan from their subscription settings, with a promo code when they have one. You can finish a project at the end of the month without changing plans.
- Model
Ideogram V4.5, FLUX 3 Image, Grok Imagine Video 1.5 and more
Ideogram V4.5, FLUX 3 Image, Grok Imagine Pro, Seedream 5 Flash, Recraft V4.1, Meta Muse Image and, on October 3, Marigold V2 for images, and Grok Imagine Video 1.5 (plus 1.5 Lite), the Extend and Edit modes of Grok Imagine Video and Kling V3 Turbo for video, join the catalogue. Fast versions such as Recraft V4.1 Flash or Grok Imagine Video 1.5 Lite help you iterate before a final render.