WELCOME TO THE CHROMA GOONTUNE!
HUGE THANKS TO MACHINEMINDED FOR MAKING ONE OF THE BEST AND MOST STRAIGHTFORWARD WORKFLOWS TO USE WITH CHROMA GENERATION!
Chroma Goontune is a style lora focused on steering Chroma toward photorealistic generation. It has been developed with a dataset spanning both amateur and professional content to ensure a broad and balanced representation of real-world detail. Think of it as reinforcement for concepts that are already existing within the uncensored base chroma model.
Training Details
Data Composition: Mixed amateur + professional sources
Captioning: Natural language captions generated with Joycaption
Objective: Improve realism, contextual accuracy, and fine detail rendering
Key Features
Enhanced photorealistic output
Strong generalization across diverse content types
Improved alignment between prompts and visual results
Use Cases
Realistic XXX NSFW Generation
Photographic-style compositions
Experimental work where lifelike textures and lighting are important
USAGE TIPS
Just a fair warning, I've found that this LoRA looks best with the workflow that I've been using. It may work better with others/different settings, so feel free to experiment and let me know!
Usage
Dependencies used in WF:
Step 1 — Find a Reference Image
Locate any image online that matches the style or subject matter you want to guide Chroma toward.
Step 2 — Generate a Caption with Joycaption
Go to a HF space like Joycaption Beta. Or use it locally if you have it downloaded.
Input the following exact prompt in the conditioning box for the output prompt:
Write a medium-length straightforward caption for this image. Begin with the main subject and medium. Mention pivotal elements—people, objects, scenery—using confident, definite language. Focus on concrete details like color, shape, texture, and spatial relationships. Show how elements interact. Omit mood and speculative wording. If text is present, quote it exactly. Note any watermarks, signatures, or compression artifacts. Never mention what's absent, resolution, or unobservable details. Vary your sentence structure and keep the description concise, without starting with “This image is…” or similar phrasing. Include information about lighting. Include information about camera angle. If applicable, mention the likely use of artificial or natural lighting sources. ONLY describe the most important elements of the image. Do NOT use polite euphemisms—lean into blunt, casual phrasing.
Step 3 — Use the Caption in Chroma Goontune
Take the generated caption and feed it into Chroma Goontune as your prompt or conditioning text.
Adjust sampling parameters as usual to refine results. You can usually switch between "an amateur candid photo" and "a professional photo" to adjust the look of the generation. Different conditionings in the prompt such as "off-angle, blurry" can lead to better amateur images.
Example Workflow:
Find a reference photo of whatever you want to try to replicate.
Run it through Joycaption with the prompt above.
Copy the generated caption (e.g., “A dimly lit urban street with neon signage above storefronts, a parked silver sedan reflecting blue light, and pedestrians crossing under overhead cables. Strong artificial light sources cast sharp shadows across wet pavement.”)
Paste the caption into Chroma Goontune using the workflow provided in the showcase images.
DISCLAIMER
Chroma Goontune is capable of producing highly photorealistic imagery. Please use this LoRA responsibly.
Do not use it to generate misleading or deceptive content.
Avoid creating imagery that could cause harm, violate personal privacy, or impersonate real individuals without consent.
Remember that outputs may closely resemble real photography — always make it clear when images are synthetic.
This LoRA is released for creative purposes only. I'm not responsible for misuse or any downstream applications that result in harm or violation of laws or ethical guidelines.
Also, I don't think I NEED to say this, but goontune.com is not a real website lol.
Description
EXPERIMENTAL VERSION OF THE LORA, IT MIGHT SUCK. IT GIVES REALLY HIGH FIDELITY PICTURES, BUT GOES INTO FLUX SLOP FROM TIME TO TIME. NEED TO TEST SOMETHING. PLEASE USE THE WF IN THE TRAINING DATA!
DEPENDENCIES NEEDED:
FAQ
Comments (26)
amazing, this is finally maximizing the potential that lodestone built into chroma
Training on Chroma-2K-QC is gold. People are going to be blown away once Chroma's community efforts mature.
It looks like the Hyper-Chroma-low-step-LoRA is not available anymore.
I noticed that a bit earlier myself… an updated version of the Lora and hopefully a new workflow will be coming very soon.
@xG00N3Rx There are some LoRA backups here: https://huggingface.co/Vinzou/Chroma-LoRA-Experiments
@roberto_baggio Nice find, thank you for sharing.
@roberto_baggio Which one would you recommend using?
@roberto_baggio Ohhh interesting. Does combining multiple low step loras yield better results? I assume the Hyper lora and the flash loras link that you shared have to be with low strength too?
For those who still need silveroxides loRAs, the previously linked repo has moved and now includes more updated files.
https://huggingface.co/Vinzou/Chroma-LoRA-Experiments
Thank you to Sliveroxides, Mcsamaster, and qpqpqpqpqpqp for their contribution in mirroring these files.
really nice work. however i seem to get fluxified skin a lot.. any way to dial it in more to look like your example images?
This is great thank you for your work. For training a character on this - would we just train a lora on chroma?
Would you mind uploading this to tensor? Or giving me permission to do so for you? I’d quite like to use it there. If you do upload it yourself, be sure to only use SFW images as a preview.
Great Lora and great workflow! Unfortunately several Loras are not available anymore. I found some alternatives and after testing can recommend the following for very realistic results:
Diffusion Model: Chroma Unlocked V46 Detail Calibrated
Lora 1: hyper-low-step-LoRa @0.25
Lora 2: hyper_flash_heun_low_step_r04_fp32 @0.25
Lora 3: 1518goontunerank64prodigy @0.70
Sampler: res_multistep
Scheduler: sgm_uniform or beta
Steps: 15
CFG: 1.5
Batch size: 4
how are you even getting it to load? This is the only Chroma LORA that won't work for me. It's giving me errors like lora key not loaded: lora_transformer_transformer_blocks_9_ff_net_2.lora_down.weight
what do all these rank loras actually do?
@CapAndABull I am sorry, I don't know why it wouldn't load for you. I tried to be specific as to the models and loras I used and they work together. Maybe there is an incompatibility with ComfyUI or a custom node.
@gilak7758 It is my understanding that they make generation faster since Chroma itself isn't really on the fast side of things.
@CapAndABull I do get similar errors, a bunch of them actually. The workflow still produces great results. I think you can ignore these errors and just get creative.
@gilak7758 To be honest. I don't know.
LORA doesn't work for me at all. Getting "key missing" and shit.
EDIT: This is literally the only LORA of all the Chroma LORA I've tried that just plain doesn't work and consistently gives me the lora key not loaded error. It usually means this was trained for a different architecture or something. Has anyone here ACTUALLY gotten it to work?
I get the same error so it isn't just you...
@dfree3305149 I have tried different versions of Chroma, including some talked about in the instructions, literally nothing works. I haven't been able to get this working with ANY workflow.
same :(
Trust me. 2K-qc has some amazing effects.
When I use the sde sampler, it stabilizes the fingers in the image earlier across multiple seeds. And across multiple prompt sets, adding a LORA with strength 1 doesn't drastically alter the composition (which is a good thing for me).
However, there's one negative change for me: most of the time, it makes skin details more “regularized” (very similar to flux.dev), and it also has a bias toward lighter skin tones in the crotch area.
I think if you adjust the Caption method or tweak the proportion of “hands” at different angles in the dataset, it could become the first LORA in Chroma with a high probability of fixing hands.
I use it without the workflow and without following the steps, I just plant it in my normal workflow and it literally makes things look better.