New Image models in ComfyUI
In this video, I test several new ComfyUI image generation workflows and models directly on my local computer without using APIs. Instead of forcing every model to use the same prompt, I create prompts that are designed specifically for each workflow, because every model has different strengths.
We look at anime-style generation, Ideogram JSON prompting, Image Lens, the P model workflow, HiDream image reference, Pixel Edit, Ernie Image Turbo, Ernie text-to-image, and FireRed image analysis. I also talk about model loading speed, SSD performance, VRAM usage, upscaling, text encoders, negative prompts, image references, and which workflows feel useful for real production.
Some models are better for t-shirt and print designs, some are great for landscapes and flowers, some work better for complex comic-style layouts, and some are useful for reference-based image creation. I also test how some generated images can be used to create 3D models.
If you are using ComfyUI for AI art, image generation, print designs, concept art, or experimental workflows, this video should help you decide which models are worth testing.
Let me know in the comments if I missed a setting, used something incorrectly, or if you found better results with these workflows. We all learn from each other.
Hashtags:
#ComfyUI #AIArt #ImageGeneration #GenerativeAI #AITools #StableDiffusion #AIWorkflow #TextToImage #AIModels #DigitalArt #ComfyUITutorial #AIImageGeneration
⚠️ AFFILIATE DISCLOSURE: We may earn a commission from purchases made through the links below at no extra cost to you.
My recommendations:
Hitem3d: https://www.hitem3d.ai/?utm_source=af...
My Patreon webpage - / geekatplay
Tutorials and packs - https://gumroad.com/geekatplay
Tutorials website - https://www.geekatplay.com
Photography - https://www.chopinephotography.com
YouTube Chapters:
00:00 New ComfyUI workflows and image models
00:07 Why comparing models with the same prompt is not always fair
00:23 Designing prompts specifically for each model
00:42 ComfyUI version and built-in workflow templates
00:54 Using local workflows without APIs
01:08 Models selected for this test
01:33 Missing models and text encoders
01:53 Starting with the anime workflow
02:01 Why the Qwen text model is powerful for complex prompts
02:32 Handling longer and more detailed prompts
02:58 Testing contextual lighting and detailed prompts
03:12 Loading speed and generation performance
03:29 Fast generation with the anime model
04:07 Why this anime model is great for printing
04:24 Turning AI images into 3D models
04:49 Creating a 3D model from one image
05:23 Reviewing the generated 3D model result
05:54 Testing the Ideogram JSON workflow
06:10 How JSON prompting works
06:26 Creating complex comic book layouts
06:59 Why JSON is powerful for detailed compositions
07:24 How to use ChatGPT or Gemini to create JSON prompts
08:00 Final thoughts on the Ideogram JSON workflow
08:29 Testing Image Lens workflow
08:46 Best use cases for Image Lens
09:02 Landscape, flowers, depth of field, and detail quality
09:27 Aspect ratio and megapixel controls
09:42 Built-in upscaling options
10:06 Why Image Lens works well for beautiful detailed images
10:21 Testing the P model workflow
10:28 Fast generation with Z Turbo and upscaling
11:09 SSD recommendation for large AI models
11:34 Quality and speed results from the P model
11:59 Testing text and detail quality
12:14 Upscaling with added details
12:31 Testing HiDream with image reference
12:54 Using HiDream for semi face-reference results
13:10 Important setting for image reference mode
13:26 Reviewing the HiDream result
14:11 Testing Pixel Edit workflow
14:28 Simple workflow structure and model setup
14:43 Using negative prompts
15:00 Texture and detail limitations
15:31 Testing Ernie Image Turbo and text-to-image
15:47 Mistral text encoder and complex prompt handling
16:40 VRAM and model loading challenges
17:04 Ernie Turbo vs non-Turbo results
17:44 Share your settings and suggestions
18:23 Testing FireRed image analysis workflow
18:40 Using reference pose and style without ControlNet
18:56 Adding multiple image references
19:10 Loading time and performance notes
19:25 Creating armor, style changes, and pose references
19:49 Final thoughts and viewer feedback request
20:00 Closing