Browse Models
Launch a dedicated cloud GPU server running Laboratory OS to download and run Photon using any compatible app or framework.
Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on local system resources, particularly GPU(s) and available VRAM.
Generate images and videos using a powerful low-level workflow graph builder - the fastest, most flexible, and most advanced visual generation UI.
Forge is a platform built on top of Stable Diffusion WebUI to make development easier, optimize resource management, speed up inference, and study experimental features.
Automatic1111's legendary web UI for Stable Diffusion, the most comprehensive and full-featured AI image generation application in existence.
Train your own LoRAs and finetunes for Stable Diffusion and Flux using this popular GUI for the Kohya trainers.
Photon is a checkpoint merge based on Stable Diffusion 1.5, developed by Photographer for generating photorealistic images. The model integrates various training adaptations and LORA modules to enhance image quality with minimal prompting. It functions effectively as an image refiner and supports further customization through LORA-based training, though limitations in anatomical accuracy persist.
Photon is a generative artificial intelligence model designed for the creation and enhancement of photorealistic images. Developed as a checkpoint merge based on the Stable Diffusion 1.5 framework, Photon integrates various training adaptations and model refinements to deliver outputs reflecting its training objectives with minimal user prompting. The model is noted for its image refinement abilities, versatility in generative tasks, and suitability for further tuning using low-rank adaptation methods.
Photon's architecture is rooted in the Stable Diffusion 1.5 latent diffusion model, a widely adopted open-source text-to-image generation system. The Photon checkpoint is classified as a checkpoint merge, having drawn from an assortment of prior model versions and fine-tuned LORA modules, each tailored to specific visual attributes and subject matter.
During its development, the model’s creator—operating under the pseudonym "Photographer"—employed a chaotic process, first merging earlier trained models, then iteratively training LORA adapters on AI-generated, photorealistic datasets. These LORA modules were integrated back into the model using dynamically weighted blending strategies to address specific representational shortcomings, particularly in the depiction of hands. While some resolution was achieved, limitations in anatomical accuracy persisted in initial releases.
The core refinements in Photon are driven by low-rank adaptation (LORA) methods that facilitate efficient retraining and merging across thematic domains. Much of the training involved curating and leveraging AI-generated photorealistic images rather than employing large-scale, human-annotated collections.
The project's stated ambition was to scale the training dataset to include between 5,000 and 50,000 high-quality, AI-crafted photorealistic samples, with the ultimate goal of further automating the refinement and blending process. This approach emphasizes adaptability, making Photon particularly effective for developing new custom LORA modules tailored to specific stylistic or semantic requirements.
Photon’s principal function is the generation of photorealistic imagery. Its outputs exhibit characteristics of photorealism, though users note a distinction between near-photorealism and photographic fidelity. The model functions as a refiner, capable of transforming certain types of visually unrefined images into outputs consistent with photorealistic styles. This refinement occurs with minimal reliance on complex or verbose prompting.
Photon is also recognized for its robust performance in pseudo image-to-image (IMGtoIMG) tasks. In these contexts, the model consistently produces realistic outputs when low denoising settings are selected, while high redrawing intensities may introduce a stylized, two-dimensional effect inconsistent with photorealism.
Another notable feature of Photon is its compatibility with further LORA-based training, making it suitable for users seeking to customize or extend its capabilities for niche visual effects or subject domains.
Photon serves several primary uses in the field of generative AI. It is frequently utilized to produce photorealistic images based on textual descriptions for content creation. The model’s image-refining capability is applied in post-processing pipelines, where AI-generated imagery can be transformed into compositions reflecting photorealistic characteristics without extensive manual intervention.
Additionally, Photon is widely employed as a foundation for LORA training, enabling targeted fine-tuning by researchers and artists. Its capabilities in image-to-image workflows allow for the realistic transformation or enhancement of existing images, contributing to its applicability within digital media and content creation domains.
Since its publication on June 5, 2023, the model has been used for image generation. The model file, in fp16 pruned format, occupies 1.99 GB, reflecting its modeling scale.
Photon exhibits versatility, responsiveness to prompt variations, and effective refinement performance. However, certain limitations have been documented. Users have reported persistent challenges with accurate hand generation, despite ongoing efforts to address these anatomical issues through iterative LORA mixing. When employing high redrawing (denoising) settings in the image-to-image pipeline, results may skew towards a flat, two-dimensional aesthetic, eroding photorealistic qualities.
While the license for the model aligns with the CreativeML Open RAIL-M standard, an addendum provides additional guidance on redistribution and responsible use.
Photon is released under the CreativeML Open RAIL-M License, which mandates responsible and ethical application of the model and its derivatives. Users are encouraged to consult the accompanying license addendum for detailed stipulations on permissible and restricted uses.