<h2 id="this-is-the-official-model-page-for">This is the official model page for</h2><h2 id="chroma-dc-2k-safetensor-and-gguf-quants">Chroma-DC-2K Safetensor and GGUF Quants</h2><h2 id="this-model-is-a-merge-of-v48-detail-calibrated-and-an-experiment-named-2k-test">This model is a merge of v48-detail-calibrated and an experiment named 2k-test</h2><h2 id="comfyui-workflow-for-the-leftmost-showcase-image-is-in-files-marked-as-training-data">ComfyUI workflow for the leftmost showcase image is in files marked as training data</h2>The following is a copy of original model page.<h1 id="hey-everyone"><a target="_blank" rel="ugc" href="https://fictional.ai/?ref=chroma_civit"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/7ab454f0-c2e6-4f1e-a34f-58a1fd99f845/width=525/7ab454f0-c2e6-4f1e-a34f-58a1fd99f845.jpeg" /></a> Hey everyone,</h1>A while back, I posted about Chroma, my work-in-progress, open-source foundational model. I got a ton of great feedback, and I'm excited to announce that the base model training is finally complete, and the whole family of models is now ready for you to use!A quick refresher on the promise here: these are true base models.I haven't done any aesthetic tuning or used post-training stuff like DPO. They are raw, powerful, and designed to be the perfect, neutral starting point for you to fine-tune. We did the heavy lifting so you don't have to.And by heavy lifting, I mean about 105,000 H100 hours of compute. All that GPU time went into packing these models with a massive data distribution, which should make fine-tuning on top of them a breeze.As promised, everything is fully Apache 2.0 licensed—no gatekeeping.TL;DR:Release branch:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Chroma1-Base">Chroma1-Base</a>: This is the core 512x512 model. It's a solid, all-around foundation for pretty much any creative project. You might want to use this one if you’re planning to fine-tune it for longer and then only train high res at the end of the epochs to make it converge faster.</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Chroma1-HD">Chroma1-HD</a>: This is the high-res fine-tune of the Chroma1-Base at a 1024x1024 resolution. If you're looking to do a quick fine-tune or LoRA for high-res, this is your starting point.</li></ul>Research Branch:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Chroma1-Flash">Chroma1-Flash</a>: A fine-tuned version of the Chroma1-Base I made to find the best way to make these flow matching models faster. This is technically an experimental result to figure out how to train a fast model without utilizing any GAN-based training. The delta weights can be applied to any Chroma version to make it faster (just make sure to adjust the strength).</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/chroma-debug-development-only/tree/main/radiance">Chroma1-Radiance [WIP]</a>: A radical tuned version of the Chroma1-Base where the model is now a pixel space model which technically should not suffer from the VAE compression artifacts.</li></ul><h1 id="quantization-options">Quantization options</h1><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Clybius/Chroma-fp8-scaled">Alternative option: FP8 Scaled Quant</a> (Format used by ComfyUI with possible inference speed increase)</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/silveroxides/Chroma-GGUF">Alternative option: GGUF Quantized</a> (You will need to install ComfyUI-GGUF custom node)</li></ul><h1 id="special-thanks">Special Thanks</h1>A massive thank you to the supporters who make this project possible.<ul><li>Anonymous donor whose incredible generosity funded the pretraining run and data collections. Your support has been transformative for open-source AI.</li><li><a target="_blank" rel="ugc" href="http://Fictional.ai">Fictional.ai</a> for their fantastic support and for helping push the boundaries of open-source AI.</li></ul><h1 id="support-this-project!">Support this project!</h1><a target="_blank" rel="ugc" href="https://ko-fi.com/lodestonerock/">https://ko-fi.com/lodestonerock/</a>BTC address: bc1qahn97gm03csxeqs7f4avdwecahdj4mcp9dytnj ETH address: 0x679C0C419E949d8f3515a255cE675A1c4D92A3d7my discord: <a target="_blank" rel="ugc" href="http://discord.gg/SQVcWVbqKx">discord.gg/SQVcWVbqKx</a>

Chroma-DC-2K

chromaDC2K_mainSafetensors_trainingData.zip

chromaDC2K_mainSafetensors.safetensors

Main Safetensors

fp8_scaled_hybrid_rev2

fp8_scaled_hybrid

Q8_0

This is the official model page forChroma-DC-2K Safetensor and GGUF QuantsThis model is a merge of v48-detail-calibrated and an experiment named 2k-test<ul><li>Try Res_Multistep sampler /Beta Scheduler </li><li>For Most prompts 35 steps is enough. For Certain complicated prompts and ones with loras, try 60 steps</li><li>This version can go resolutions nearing 2K. The max tried possible in Tensor Art is 1750x1312</li></ul>Try this negative if you keep getting cartoon/anime output and want realism:bad anatomy. malformed. Deformed hands. deformed fingers. blurry. 4 fingers. 6 fingers. fused fingers. long fingers. 3d, anime, cgi, drawing, cartoon drawing, ai-generated, pixelated. anime, doll skin, , illustration, sketch, painting, perfect skin, 2d. 2.5d. smooth plastic skin, airbrushed, soft focus, plastic skin, unnatural skin texture, digital painting.The following is a copy of original model page.Hey everyone,A while back, I posted about Chroma, my work-in-progress, open-source foundational model. I got a ton of great feedback, and I'm excited to announce that the base model training is finally complete, and the whole family of models is now ready for you to use!A quick refresher on the promise here: these are true base models.I haven't done any aesthetic tuning or used post-training stuff like DPO. They are raw, powerful, and designed to be the perfect, neutral starting point for you to fine-tune. We did the heavy lifting so you don't have to.And by heavy lifting, I mean about 105,000 H100 hours of compute. All that GPU time went into packing these models with a massive data distribution, which should make fine-tuning on top of them a breeze.As promised, everything is fully Apache 2.0 licensed—no gatekeeping.TL;DR:Release branch:<ul><li>Chroma1-Base: This is the core 512x512 model. It's a solid, all-around foundation for pretty much any creative project. You might want to use this one if you’re planning to fine-tune it for longer and then only train high res at the end of the epochs to make it converge faster.</li><li>Chroma1-HD: This is the high-res fine-tune of the Chroma1-Base at a 1024x1024 resolution. If you're looking to do a quick fine-tune or LoRA for high-res, this is your starting point.</li></ul>Research Branch:<ul><li>Chroma1-Flash: A fine-tuned version of the Chroma1-Base I made to find the best way to make these flow matching models faster. This is technically an experimental result to figure out how to train a fast model without utilizing any GAN-based training. The delta weights can be applied to any Chroma version to make it faster (just make sure to adjust the strength).</li><li>Chroma1-Radiance [WIP]: A radical tuned version of the Chroma1-Base where the model is now a pixel space model which technically should not suffer from the VAE compression artifacts</li></ul>Special ThanksA massive thank you to the supporters who make this project possible.<ul><li>Anonymous donor whose incredible generosity funded the pretraining run and data collections. Your support has been transformative for open-source AI.</li><li>for their fantastic support and for helping push the boundaries of open-source AI.</li><li></li></ul>Support this project!https://ko-fi.com/lodestonerock/BTC address: bc1qahn97gm03csxeqs7f4avdwecahdj4mcp9dytnjETH address: 0x679C0C419E949d8f3515a255cE675A1c4D92A3d7my discord: discord.gg/SQVcWVbqKx

Chroma-DC-2K.safetensors

This is the official model page forChroma-DC-2K Safetensor and GGUF QuantsThis model is a merge of v48-detail-calibrated and an experiment named 2k-test

Files with SHA256: 4face87605d8a5b7e8c020f6d67426f60e11f345236738edfc56ecf739ee65ba