Model update · July 14, 2026
Four new AI model workflows, and a clearer way to choose between them
Seedream 5, HappyHorse 1.1, and Seedance 2.0 Mini are now connected to live generation on Voor. We also published a Kling 3.0 Turbo workflow page that keeps the currently available Kling v3 model labeled honestly.
A real Seedream 5 example, used here as the starting point for the new model directory.
Seedream 5
Reasoning-led image generation for structured compositions, material constraints, short readable copy, and controlled edits.
Open workflowHappyHorse 1.1
Text-to-video, image-to-video, and reference-to-video for dialogue, narrative beats, and native-audio work.
Open workflowSeedance 2.0 Mini
A smaller Seedance lane for social hooks, product loops, prompt tests, and reference-led video drafts.
Open workflowKling 3.0 Turbo workflow
A production guide for Turbo-style shot planning. The generator remains clearly labeled Kling v3 until the Turbo API is connected.
Open workflowSeedream 5: choose Lite for iteration, Pro for the demanding frame
Seedream 5 is the most substantial image update in this release. Both Lite and Pro sit in one family selector, so a brief can move between them without being rewritten. Lite is the natural first pass for composition, hierarchy, and prompt debugging. Pro is the finishing lane when material detail, complex relationships, or a high-value hero image justify the heavier run.
The useful change is not a longer adjective list. Seedream 5 works best when a brief exposes its rules: what information must be correct, what objects must relate to one another, which text must remain readable, and what can be stylized. Treat every generated fact or label as a draft and verify it before publishing.
HappyHorse 1.1: one family for dialogue and reference video
HappyHorse 1.1keeps three distinct production modes together: start from text, animate a first frame, or guide the result with references. That makes it a better fit for short narrative work than a generic “make this image move” prompt.
Write spoken lines exactly, keep each action physically ordered, and name the role of every reference. A face reference should not silently redefine the set; a motion reference should not replace wardrobe. Native audio is useful, but dialogue timing, room tone, and commercial rights still need a human review.
Seedance 2.0 Mini: optimize for decisions, not clip count
Seedance 2.0 Miniis the fast lane for testing a hook, turntable, reaction beat, or five-second product loop. It supports text, image, and reference starts, but its real advantage is the cost and speed of reaching a clear yes-or-no decision.
Change one variable per take, compare variants side by side, and promote only the winner into a heavier finishing model. If eight drafts do not answer a written creative question, the prompt is still too broad.
Kling 3.0 Turbo: the page is ready before the model is
The Kling 3.0 Turbopage is deliberately different. It includes real Turbo examples and a practical shot-planning workflow, while the live generator is still labeled as the available Kling v3 model. We do not rename a standard v3 render to make an unavailable API look connected.
You can still prepare production-ready briefs now: one subject, one camera move, one action, one finish condition, and a frame-level reject list. When the verified Turbo schema is available, those same briefs can move over without a misleading comparison.
Also new: focused tools for icons and image translation
- App Icon Generator turns a product brief into crop-safe icon concepts with Recraft v3 and packages the selected artwork for common platforms.
- AI Image Translator uses GPT Image 2 Edit to replace visible text while protecting the original image, hierarchy, and layout.
Pick a workflow