Segmind Release Notes (October 5–11, 2026)

1. New Models — Image, Text and Upscaling

• What's new: Several models are now live on the API: FLUX 3 Image (flux-3-text-to-image and flux-3-image-edit), Nano Banana 2.1, Gemini 3.8 Flash, Qwen-Image 2.1 Pro, and Kandinsky 6 VSR, a video upscaler with Pro and Lite modes. Kandinsky 6 VSR shows its per-request cost on the model page.
• Impact: Generate and edit images with the latest FLUX and Qwen models, use Gemini 3.8 Flash for fast text and multimodal tasks, and upscale video at the quality-versus-speed level you choose.
• Released: October 6–7, 2026

2. Grok Imagine Video 1.5 — Reference Audio, Last Frame and Keyframes

• What's new: grok-imagine-video-1.5 now accepts reference audio, a last frame and keyframes.
• Impact: Direct how a video ends and how it moves between key moments, and guide its soundtrack with reference audio.
• Released: October 7, 2026

3. Gemini Video Understanding and Multimodal Input

• What's new: The new gemini-3.8-flash-video-agent model answers questions about a video file or YouTube link and returns a text answer. Separately, Gemini models now accept image, video, audio and PDF inputs on every interface, including multi-file fields, across the API, the playground and Pixelflow. Videos without an audio track are now analyzed reliably.
• Impact: Ask questions about a video, or send mixed media to a Gemini model in a single request, without converting files first.
• Released: October 8–9, 2026

4. Hunyuan 3D 3.1 Pro and the 3D Viewer

• What's new: Hunyuan 3D 3.1 Pro (hunyuan-3d-3.1-pro-image-to-3d) turns a front image, plus up to seven additional views, into a textured 3D model, returned as GLB or OBJ. 3D results now open in an interactive viewer in the playground and in Pixelflow, with orbit controls, touch hints and download, and each result includes a thumbnail image.
• Impact: Go from product photos to a textured 3D asset, then rotate and inspect it right where you generated it.
• Released: October 9–10, 2026

5. MiniMax Speech and Voice Studio

• What's new: The MiniMax Speech text-to-speech family (versions 2.8, 2.6 and 02, in HD and Turbo) is live and works with the full set of system voices. Per-character pricing is shown on the model page. In Voice Studio, you can now delete your custom voices, and HeyGen voices are available alongside the existing options.
• Impact: More choice of speech models and voices, clear per-character cost, and control over the custom voices in your library.
• Released: October 7–11, 2026

6. Pixelflow — Model Swap, Reliable Runs and Bigger Workflows

• What's new: You can swap a node's model in place, keeping its connections and compatible settings, and Pixelflow shows what the swap would change before you confirm. Nodes using a retired model are now flagged with Swap offered as the fix. Failed fan-out runs are now visible, with an option to stop on the first error. Downloading multiple files saves a single zip, and workflows larger than 1 MiB can now be saved.
• Impact: Upgrade a workflow to a newer model without rebuilding it, spot problems before a run, and keep working with large workflows.
• Released: October 6–9, 2026

7. Account, Credits and Connected Apps

• What's new: Your live credit balance and team spend limits now appear in the console header and sidebar. When you authorize an app to connect to Segmind, you can choose which account it bills on, and you can revoke each app connection individually.
• Impact: Keep track of credits at a glance and stay in control of which account and apps are using your Segmind access.
• Released: October 7–10, 2026

8. Playground, Generations and Discovery

• What's new: Add to assets in the model playground saves a generation to your asset library, with a countdown showing when it would otherwise be removed. Generations rows now show the failure reason inline, and requests with missing required inputs return a clear error that names the problem. The Models page now shows one scrolling row per modality, and the MCP page shows real example outputs and current setup steps. API requests to POST /v2/{model} can include an X-Request-ID so a lost response can be recovered by polling /v2/requests/{id} (PRO).
• Impact: Keep the generations you want, understand failures faster, find models more easily, and recover reliably from interrupted API calls.
• Released: October 6–11, 2026