Another AI model just landed.
This time, it's Tencent's Hy Image 3.5 Preview — a unified image generation and editing model that puts several workflows behind a single endpoint.
And it showed up on DEV·TV's New Models channel just 5 hours after being added.
That's exactly why I built the channel.
🖼️ One endpoint. Three modes.
Hy Image 3.5 Preview brings together:
✍️ Text → Image
Generate an image from a prompt.
🖼️ Image → Image
Give it an existing image and transform it.
🔄 Multi-turn Editing
Keep refining the result instead of starting over.
Think:
"Make the sky darker."
Then:
"Remove the car."
Then:
"Make the lighting warmer."
Same workflow. Same model.
No jumping between separate generation and editing systems.
🧩 Up to 20 reference images
This is where things get interesting.
Hy Image 3.5 Preview supports up to 20 reference images in a request.
That opens up some useful workflows:
🎨 Style consistency
🧑 Character consistency
📦 Product references
🖼️ Multiple visual sources
Instead of describing everything from scratch, you can give the model the visual material you want it to work with.
⚡ The specs
| Model | tencent/hy-image-v3.5-preview |
| Context | 100K |
| Reference images | Up to 20 |
| Generation | Text → Image |
| Editing | Image → Image |
| Multi-turn | Yes |
| Status | Preview |
DEV·TV currently shows it as free.
As always with preview models, pricing and availability can change.
🤔 Why this matters
The interesting part isn't simply "Tencent released another image model."
It's the consolidation.
A developer building an image workflow can potentially use one integration for:
Generate → Edit → Refine → Edit again
That's a much simpler mental model than stitching together separate tools for every stage.
And with a 100K context window, there's plenty of room for longer prompts and multi-turn interaction.
⚠️ It's still a preview
Hy Image 3.5 Preview isn't something I'd blindly put into production tomorrow.
Preview means:
- Rate limits can change
- Model behavior can change
- Availability can change
- Pricing can change
- The API can evolve
So experiment with it.
Build something weird.
See where it breaks.
That's half the fun of new models anyway.
📺 And this is why DEV·TV has a New Models channel
AI models are arriving too quickly to keep track of manually.
You can follow a dozen AI companies and still miss something interesting.
So DEV·TV does something simpler:
It puts the developer internet on TV.
New models.
GitHub projects.
Developer news.
Research.
Security.
Framework releases.
And now, when a new model appears in the catalog, it can get airtime.
No login.
No backend.
No API keys.
Just turn it on.
Today's model:
Tencent — Hy Image 3.5 Preview
tencent/hy-image-v3.5-preview
100K context · image generation · image editing · multi-turn · up to 20 references
Spotted on DEV·TV 5 hours after it was added.
Tomorrow, something else will probably be there.
That's the point. 📺
👉 Try DEV·TV: https://shouvik12.github.io/devtv/
⭐ If you like the idea, consider starring the DEV·TV repo on GitHub. It helps the project get discovered by more developers.
DEV·TV is free and open source — a TV for the developer internet.
Top comments (0)