FULL STORY

Tencent Hunyuan HyImage3.5: From Launch to Hands-On

Tencent Hunyuan launched HyImage3.5 preview on Sept 22, touting 2K image generation at just $0.02, followed by ComfyUI availability and hands-on reviews highlighting its text rendering and editing control.

2026-09-22 ~ 2026-09-23 · 2 episodes · 18 posts

Episode 1 · Tencent Hunyuan Launches HyImage3.5 Preview: 2K Image Generation at $0.024 Each (2026-09-22, 10 posts)

On September 22, Tencent Hunyuan officially released HyImage3.5 preview, a professional-grade yet low-cost image generation model with API pricing as low as $0.024 per image (0.15 RMB), launching simultaneously on OnSolo and ComfyUI. It marks a generational upgrade for the Hunyuan image series, drawing attention with 2K output and industry-low pricing.

Confirmed

  • HyImage3.5 preview supports text-to-image and image-to-image, up to 5 reference images per generation, multiple aspect ratios, up to 2K resolution, and multi-turn conversational editing.
  • In internal GSB blind tests with hundreds of designers, overall performance improved 30% (30% higher human-eval win rate) over HyImage3.0.
  • API price is $0.024 per image (0.15 RMB), reported by Zhidongxi as the lowest in the industry.
  • Official use cases include short-drama character sheets, dynamic game assets and keyframe generation, with characters staying consistent across episodes.
  • The model launched exclusively on OnSolo with 5 reference images, 2K output, 1 credit per generation, and free unlimited generation for members during a two-week promotion.
  • It is also available on ComfyUI, supporting text-to-image and image-to-image in one model, multilingual text and symbol rendering.
  • @HeyAmit noted the model quietly went live on Sept 22 in a limited-time free window and plans to stress-test it with harder prompts.

Why it matters

  • The ultra-low price combined with 2K output directly disrupts image-generation market pricing, called the industry's lowest by Zhidongxi.
  • Distribution spans both a third-party platform (OnSolo exclusive, free for members) and the open-source workflow ComfyUI. @Divpradeep and @LearnWithBishal both reported good quality, fast generation and competitive pricing; @HeyAmit argued such models unlock entire workflows—product photos to ad creatives, game concepts to UI, reference images to new visuals—rather than just single-image generation.

Episode 2 · Hands-on: Tencent Hunyuan Image 3.5 Preview Excels at Text and Editing Control (2026-09-23, 8 posts)

On September 23, blogger HeyAmit published a 6-post thread testing Tencent's Hunyuan Image 3.5 preview. His overall verdict: the model's value lies not just in image quality but in fine-grained control over generation and embedding it into fuller workflows. Testing was done through Miora's Image Generator, with a pipeline of prompt/reference images → generation → review.

Confirmed

  • Supports both text-to-image and image-to-image, with up to 5 reference images as visual context rather than relying on a single prompt.
  • Text generation works across languages, font sizes, and layouts, and understands how typography fits the composition—suited to posters, ads, presentations, and UI concepts per the blogger.
  • For realism, the model emphasizes lighting, materials, lens language, environmental detail, and texture, which greatly affect image credibility.
  • In editing tasks (scene replacement, style adjustment, image transformation), the design goal is preserving key features of the reference image while applying user-requested changes; in testing, changing scenes and styles retained reference features.
  • Workflow extensions cited: product photo → ad creative, idea → poster, game concept → UI, reference image → new visuals.

Why it matters

Text rendering has long been a pain point in AI image generation, while multi-reference and feature-preserving editing determine whether a model fits real production pipelines. If stable, Hy Image 3.5 could become practical for posters, ads, presentations, and UI design. As a preview release, final performance remains to be seen.