Chroma 1 Radiance Text to Image

Chroma 1 Radiance text-to-image generation in pixel space. Write a prompt and generate.

Chroma Radiance is architecturally different from most image models. Standard diffusion models encode images into a compressed latent space and then decode them through a VAE, which introduces color artifacts, blurring, and preview inaccuracy. Chroma Radiance generates directly in pixel space, skipping the latent-to-image decode step entirely. The result is sharper output, more accurate colors, and previews that match the final image.

The default prompt is a macro scene of tiny bakers working on a croissant: four paragraphs covering subject, lighting, material detail, and technical specs. That's the level of specificity the model handles without losing coherence.

How do you use Chroma 1 Radiance for text-to-image generation?

Write a prompt, set your resolution, and run. Chroma Radiance uses a BetaSampler at 30 steps and generates directly in pixel space. No VAE needed. Flow shift, CFG, and ChromaRadiance tile settings are all adjustable. The model handles long, detailed prompts accurately and is open for commercial use under Apache 2.0.

Positive prompt Chroma Radiance was built for detailed, multi-part prompts. The default prompt covers subject, action, material texture, lighting, depth of field, and resolution specification in one continuous description. That approach works.

Prompting structure that works: Lead with subject and action: "hyperrealistic macro photograph of a team of tiny bakers collaborating on a croissant." Layer material detail: name textures, surfaces, and finishes specifically. "Flaky, layered textures." "Rough wood of the worktable." "Slight sheen of melted butter." Name lighting behavior: "warm, soft kitchen lighting," "cinematic depth," "subtle highlights on the golden crust," "gentle shadows that emphasize texture." End with technical specs: "ultra-detailed, 8K resolution, photorealistic textures, sharp focus, shallow depth of field." The model follows long prompts and benefits from specificity on anatomy, hands, and complex scenes where other models drift.

Negative prompt The default negative prompt targets Chroma Radiance's known weak points: low quality, blurring, chromatic aberrations, oversaturation, and "toony" aesthetics. Leave it as-is and extend it if specific artifacts appear in your outputs.

Resolution (default: 1024x1024) 1024x1024 is the standard starting resolution. Chroma Radiance's pixel-space generation handles resolution changes cleanly without the upscale artifacts that VAE-based models can introduce.

Steps (default: 30) 30 steps via BetaSamplingScheduler. The BetaSampler with alpha and beta values at 0.4 produces stable, detailed output. Reducing steps speeds up generation; the model holds quality well down to around 20 steps for preview runs.

CFG scale (default: 3.5) Controls prompt adherence. 3.5 is calibrated for Chroma Radiance's pixel-space architecture. Higher values tighten prompt following; lower values give more interpretive freedom.

Flow shift (default: 1.0) The builder notes: "1.0 represents the intended shift by the creator and produces greater detail. Raise up to 3.0 to help with complex compositions." For standard prompts, leave at 1.0. For dense, multi-element scenes with complex spatial relationships, increasing toward 2.0-3.0 helps the model maintain structural coherence.

ChromaRadianceOptions Contains the nerf_tile_size setting. Increase this value as high as your GPU memory allows for faster generation speed. The tooltip in the node provides guidance on the limit.

Sampler Euler sampler with BetaSamplingScheduler. This combination is set specifically for Chroma Radiance's pixel-space architecture. Leave the sampler as-is unless you're experimenting with scheduler variants.

What is Chroma 1 Radiance text-to-image good for?

Chroma 1 Radiance is strongest for detailed, high-fidelity image generation where color accuracy, sharp previews, and complex prompt adherence matter. Direct pixel-space generation eliminates the VAE artifacts that affect most diffusion models. Handles anatomy, hands, and multi-element compositions better than latent-space models at comparable settings.

Complex scenes and detailed compositions. The model handles long, multi-element prompts without losing structural coherence. Describe a full scene with multiple subjects, specific material textures, and precise lighting and the model follows it. Dense compositions that break other models work here.

Commercial product and marketing visuals. Artifact-free color output, accurate previews, and Apache 2.0 licensing make Chroma Radiance a practical choice for commercial production. What you see in the preview is what you get in the final output, which reduces workflow iteration time.

Portraits, hands, and anatomy. Chroma Radiance handles faces, hands, and human proportions with higher accuracy than most comparable models. For portrait work or any subject where anatomical precision matters, it's a strong starting point.

Concept art and rapid prototyping. Fast inference with the nerf_tile_size optimization and stable quality across prompt complexity makes it useful for rapid visual exploration. Game development, comic creation, and concept art workflows benefit from the consistent output quality.

Honest notes: Chroma 1 Radiance is a pre-release model. The current version is v0.4 and newer versions in .pth format are available from the LodestoneRock HuggingFace. As a pixel-space model, VRAM usage may differ from standard latent diffusion workflows. Adjust nerf_tile_size based on your available GPU memory.

How does Chroma 1 Radiance compare to standard diffusion models?

Chroma 1 Radiance skips the VAE entirely by generating in pixel space rather than latent space. This removes color artifacts, decoder blurring, and the preview mismatch common in VAE-based models. The tradeoff is a different VRAM profile and a less established community around sampler tuning compared to mature models like Flux or SDXL.

Most diffusion models (Flux, SDXL, SD 3.5) encode into a compressed latent space, process there, and then decode back to pixels through a VAE. Each step introduces potential quality loss. Chroma Radiance operates directly in pixel space from the first step, which is why the output is sharper and previews match the final image.

For users who need color-accurate commercial output, consistent preview-to-final fidelity, and strong performance on complex prompts without model finetunes or LoRAs, Chroma Radiance is worth testing. For users who need an extensive LoRA ecosystem, established community workflows, and wide sampler compatibility, Flux and SDXL remain more mature options.

FAQ

What is Chroma 1 Radiance and how is it different from other image models?
Chroma 1 Radiance generates images directly in pixel space rather than using a latent space and VAE. This eliminates the color artifacts, blurring, and preview mismatch that VAE decoding introduces in standard diffusion models. The result is sharper images, more accurate colors, and previews that match the final output.

Does Chroma 1 Radiance need a VAE?
No. The pixel-space architecture eliminates the VAE entirely. The workflow uses a pixel_space VAE type as a placeholder in the node graph, but there is no latent-to-pixel decoding step that introduces artifacts. This is the core architectural difference from Flux, SDXL, and similar models.

What flow shift setting should I use for Chroma 1 Radiance?
The default is 1.0, which the model's creator designates as the intended setting for detailed output. Increase toward 3.0 when working with complex compositions that have many spatial elements. For standard single-subject prompts, 1.0 is the right starting point.

Can I use Chroma 1 Radiance for commercial work?
Yes. Chroma 1 Radiance is licensed under Apache 2.0, which permits commercial use and modification without restrictions.

What does the nerf_tile_size setting do in ChromaRadianceOptions?
It controls tile size for the model's internal processing. Setting it as high as your GPU memory allows increases generation speed. The exact limit depends on your hardware. The node tooltip provides guidance on what values to try.

How do I run Chroma 1 Radiance online?
You can run Chroma 1 Radiance online through Floyo. No installation, no setup. Open the workflow in your browser, write your prompt, and hit run. Free to try.