The most useful versions of GGUF(+the FP8 version)
EDIT:
*GGUF Workflow, since i can't upload the .json here: https://www.kombitz.com/2025/04/17/simple-comfyui-workflow-for-hidream-i1-gguf/
*missing nodes? run Comfy update\update_comfyui.bat
*quality comparison here: https://artificialanalysis.ai/text-to-image/model-family/hidream#quality
Lighter versions of the Full HiDream quantized by city96
download and use locally. nodes here:
https://github.com/city96/ComfyUI-GGUF
vae and text encoders here:
https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/tree/main/split_files
Description
FAQ
Comments (15)
same as flux it's impossible to draw something sharpen without DOF background
For the opensource community, Hidream is dead on arrival. Just like Flux, its too big to train/fine tune. The community has already given up on flux, the best the community can do is train lora's and merge them with Flux dev.
flux is not really open (try installing pro or check license). and the reason why it's not easy to train is more technical, not the size. that's why i promote hidream here.
We gave up on flux because of licensing not because of size.
FLux is untriable because the license, second is distilled model and is hard to train, today there are not a simple way. Hidream in other hand is not distilled and is MIT license.
@anonlinuxmint962 Check out Chroma. It is still in training, but shows a lot of promise as I see it lodestones/Chroma · Hugging Face
@StreamofStars Thats pretty sweet, but its very much targeted at the pony/furry/anime audience. Id be willing to support a realistic NSFW run.
@charuru Thats cope, FLUX.1-schnell model is released under the Apache 2.0 license. All you all just dumped on me because y'all cant handle the truth, or the truth offended y'all. That Hidream is too big to fine tune. The open source community is super toxic and entitled, they expect everything for free and knowing people will not be able to fine toon it for gooners because its too expensive upsets them a lot.
@fox23vang226 schnell does wax dolls not realistic people. hidream can be trained on a lot of hardware rented for example.
@andux I look at the feed daily, most people enjoy making wax doll gooner pics. Their prompts usually do nothing to move away from the wax doll look.
@fox23vang226 It has a lot of potential. But it is currently being (re)trained for the 1Mpix range. I am not really into "gratuitous" nudity (even though I see nothing wrong with it per se, and have been experimenting with models that support it due to their support for anatomy and capacity for spicy stuff when suitable for my story board concepts and gens) but it can do that do.
@fox23vang226 See my profile if you wish for some realistic experiemts
大小永远不是问题，兄弟
Here's the cold truth, but it's complicated and it's a lot.
With every release of a new type of diffuser model it's been growing in size and shrinking in usability because of it, but it's not the only reason. Civitai has on multiple occasions been nerfing content on the platform through both policies and gated (paid) access.
If anyone was awake during the launches of all the various models going back to SD1.5 > SDXL > Flux, etc... The downloads and activity have gone down drastically if you simply compare how many downloads each model had the first month, just to set an example that the AI-craze as a whole is cooling down.
A simple check at Steam who has more than 100.000 systems in their hardware survey shows that the average Joe is rocking an 8-12GB GPU, which every model since SDXL has failed to target.
We can also observe communities like StabilityAI, UnstableDiffusion and MidJourney has nearly become a graveyard compared to what they was back during SD1.5 and SDXL.
TLRD:
1. The demand for models is going down because nobody on a normal household budget can afford a 16-22GB GPU these days.
2. The supply of models are also going down because of licenses holding model-creators back, e.g. SD3 launch, and Flux.
3. Civitai is constantly worsening their service on a regular basis, making people leave or avoid the platform.
4. The AI-craze has been in steady decline for years, the SD1.5 days are long gone.
@punkbuzter340 IS SO OVER LOCAL FRIENDS