Text and reference input
Describe a new scene with a text prompt, or upload a reference image and explain the changes you want.
30:00:00
Limited OfferLimited Offer
Get 50% OFF on Annual PlansGet 50% OFF on Annual Plans→Claim 50% OffTencent Hunyuan · Preview · Generate + Edit
Hy Image 3.5 Image Generation & Editing
Hy Image 3.5 is Tencent Hunyuan's chat-style image model, built to generate a picture and then keep editing it across turns. This page covers what the preview documents, what changed from 3.0, and the live workspace where you can start creating now.
Fixed credits shown before submission, based on resolution and references. Failed tasks are refunded. Closing this page does not cancel generation.
Creative workspaceReady for your next idea
One brief. A new visual direction.
Create your first image, combine references, or describe a focused edit. Your results appear here.
Product photos•Campaigns•Reference edits
Model guide
What is Hy Image 3.5?
Hy Image 3.5 is Tencent Hunyuan's image model for creating pictures from text and refining reference images. Build a poster, restage a product or carry one character across different scenes. Multi-turn editing lets you keep the parts you like and change the background, lighting or composition step by step.
Read the official Tencent Cloud guide
Text-to-image and reference input
Documented multi-turn editing
Explicit size up to 4096 x 4096
Seed and area controls
Documented capabilities
What the Hy Image 3.5 preview can do
Explore text-to-image creation, reference edits and multi-turn workflows described in Tencent's official guide.
Multi-turn editing
Refine an image across several turns. Keep the details you like and adjust the subject, background or lighting one step at a time.
Search-enhanced rewriting
Search-enhanced prompt rewriting can expand a short brief with more context before rendering.
Dimension and seed control
Set an explicit width x height, pick an area preset (1K, 1.5K, 2K) or pass a fixed size up to 4096 x 4096, and reuse a seed for repeatable experiments.
Copyable briefs
Hy Image 3.5 prompts for real work
Each card names the goal, the mode and the parts a reviewer should check. Click Use this brief to load the prompt into the workspace above.
Case 01 · Text to image
Event poster with exact headline text
Give the model the exact words in quotation marks, state the hierarchy, then keep every other instruction visual. Text-led posters are a good way to judge how a prompt handles short, high-contrast lines.
Create a vertical event poster for a trail race. Exact headline in large condensed capitals: "FIRST LIGHT RUN". Below it, smaller: "5 OCT · DESERT RIDGE · 6 AM". A small runner silhouette on a ridge at sunrise, warm crimson and sand palette, clean grid, generous empty space, no other words, no logos, no watermark.
Case 02 · Image to image
Product scene edit with the item preserved
Upload the product, then name the parts that must not change. For a running shoe, specify that its shape, mesh, sole and laces should stay the same while the scene changes. Compare the uploaded file against the result at full size before publishing.
Keep the uploaded running shoe exactly as it is. Place it on a wet basalt rock above a mountain lake at sunrise, add drifting mist, keep the mesh, sole and laces unchanged, photorealistic product campaign, shallow depth of field, no text.
Case 03 · Multi-turn consistency
Keep one character across turns
Keep the character recognizable as the scene changes. Name the face, hair and clothing that must stay consistent, change one detail per turn, and reuse the previous result as a reference.
Keep the same fictional courier character: olive field jacket, short dark hair, small scar above the left eyebrow. Show the character in a rain-soaked alley at night, then describe the same face with the same jacket in a sunlit desert bus stop for the next shot. Consistent identity, cinematic still, no text.
Four-step workflow
How to use Hy Image 3.5 style editing today
Write the brief
Name the subject, the scene, the lighting and the deliverable. Put exact text in quotation marks and list the parts that must stay unchanged.
Pick a mode
Use text to image for a new visual. Switch to Edit images when you want to restage an uploaded product, room or portrait while keeping its details.
Set size and cost
Choose an aspect ratio and resolution. The workspace quotes a fixed credit amount before submission so there is no surprise charge.
Inspect and iterate
Compare the result with the source, zoom into text and edges, then use Edit this to feed the finished image into the next brief.
Want a deeper walkthrough with three full cases, failure points and copyable prompts? Read Hy Image 3.5: what's new, prompts and how to use it.
Commercial comparison
Hy Image 3.5 vs Hy Image 3.0 vs GPT Image 2.5
Compare input methods, editing workflows and output sizes for the job you need to finish.
| Decision | Hy Image 3.5 | Hy Image 3.0 | GPT Image 2.5 |
|---|---|---|---|
| Documented model | Hy-Image-3.5-Preview (hy-image-v3.5-preview) | Hy-Image-3.0 (hy-image-v3) | GPT Image 2.5 (Flare / Sunburst) |
| Workflow | Create an image and refine it across multiple turns | Generate from a single prompt | Prompt plus optional reference images |
| Editing | Documented multi-turn editing and reference images | Reference-to-image generation | Edit images with references in one workspace |
| Resolution | Area preset up to 2K; explicit size up to 4096 x 4096 | Width and height 512-2048, area up to 1024 x 1024 | 1K, 2K and 4K in this workspace |
Choose Hy Image 3.5
When you want to refine a scene across several turns while keeping the subject and composition consistent.
Stay on Hy Image 3.0
When a single prompt with 512-2048 pixel sides fits your image workflow.
Use GPT Image 2.5 here
When you want to generate or edit in the browser now, with aspect ratio and 1K/2K/4K controls, a fixed credit quote and recoverable tasks.
Why create on Banana Pro AI
One workspace instead of five provider accounts
Choose your model
Choose GPT Image 2.5 Flare or Sunburst in the workspace, then set the aspect ratio and resolution for your image.
Continue from a result
Edit this sends a finished image back in as a reference so you can keep iterating without a separate upload step.
Download and recover
Download a finished image or return after a refresh to check your saved task.
Hy Image 3.5 FAQ
Frequently asked questions
What is Hy Image 3.5?
Hy Image 3.5 is Tencent Hunyuan's image model for text-to-image creation, reference-based edits and multi-turn refinement. Use it for posters, product scenes and consistent characters.
Is Hy Image 3.5 the same as Hunyuan Image 3.5?
Yes. "Hy Image 3.5", "hunyuan image 3.5" and "hy-image-3.5" all point to the same Tencent Hunyuan model line. This page uses Hy Image 3.5 as the short name.
Does Hy Image 3.5 support image-to-image and multi-turn editing?
Yes. Hy Image 3.5 supports reference images and multi-turn editing. Keep the subject details consistent and change one part of the image at a time.
What resolution does Hy Image 3.5 produce?
Tencent documents area presets up to 2K through generate_max_pixels and explicit dimensions up to 4096 x 4096.
How many credits does an image cost?
The workspace shows the credit cost for your selected resolution and reference images before you generate. Failed tasks are refunded through the task system.
Can I keep editing a result?
Yes. Use Edit this to send a finished image back into the workspace as a reference, then describe the change you want. Keep the details you like and refine one part at a time.
Create an image from a clear brief
Start with one subject, one delivery format and the details that must survive. The workspace shows the model and the credit quote before anything runs.
Create an image