|
--- |
|
pipeline_tag: text-to-image |
|
widget: |
|
- text: >- |
|
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, photorealistic, ocean, |
|
output: |
|
url: ComfyUI_temp_dxhdq_00573_.png |
|
- text: >- |
|
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, powerful aura, imposing, anime style, 1 guy, cast in shadow, red glowing eyes, manic smile |
|
output: |
|
url: ComfyUI_temp_bqhxr_00036_.png |
|
- text: >- |
|
A dark, moody portrait of the holy mary juggling spheres, sacred geometry, dark background, golden ratio composition, hyper realistic, high resolution, photography, in the style of Roberto Ferri and Emil Melmoth |
|
output: |
|
url: ComfyUI_temp_bqhxr_00022_.png |
|
- text: >- |
|
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, Vegeta, iconic Saiyan prince from DBZ, (powerful stance:1.3), (muscle definition:1.2), in mid-battle roar, (Super Saiyan transformation:1.5), crackling aura of energy enveloping him, dynamic background showcasing a devastated battlefield reminiscent of Namek or Earth during epic confrontations; elements of Akira Toriyama's signature art style blended seamlessly with high saturation and bold lines to capture the intensity and raw power synonymous with Dragon Ball Z; dramatic lighting casting strong shadows to enhance Vegeta's chiseled features and battle torn armor; camera angle low and looking up to emphasize his dominance and unyielding spirit. |
|
output: |
|
url: ComfyUI_temp_bqhxr_00052_.png |
|
- text: >- |
|
tiger |
|
output: |
|
url: tiger.png |
|
- text: >- |
|
The image features an older man, a long white beard and mustache, He has a stern expression, giving the impression of a wise and experienced individual. The mans beard and mustache are prominent, adding to his distinguished appearance. The close-up shot of the mans face emphasizes his facial features and the intensity of his gaze. |
|
output: |
|
url: oldman.png |
|
- text: >- |
|
score_9, Side View of a Roman Warrior pierced By a spear, cinimatic |
|
output: |
|
url: GIlhXbZWgAAAVdi.jpeg |
|
license: gpl-3.0 |
|
--- |
|
|
|
<Gallery /> |
|
|
|
|
|
## Introducing Proteus-RunDiffusion: A New Era in Artistic AI |
|
|
|
|
|
Proteus-RunDiffusion heralds a new era in AI art creation, evolving from the pioneering Pony Diffusion v6 XL CLIP. We've overhauled the CLIP architecture, infusing it with a unique retraining approach that not only preserves character recognition and boosts NLP capabilities but also maximizes the creative scope of AI-generated art. This advancement not only builds upon Pony's groundwork by significantly enhancing its style versatility and accuracy but also integrates Proteus prompt comprehension, further extending Pony's stylistic capabilities. The introduction of Proteus-RunDiffusion offers artists an unmatched tool, enabling the production of diverse, high-quality artworks with remarkable precision, while ensuring SDXL model compatibility and intellectual property respect. This reimagined CLIP model propels AI artistry forward, empowering creators to explore new artistic horizons and realize their most innovative ideas. |
|
|
|
https://rundiffusion.com/proteus-rundiffusion#view-generation-samples |
|
|
|
key acheivement: |
|
- |
|
The thorough redesign and complete retraining of CLIP has resulted in an innovative phenomenon termed "style unlocking". This enhancement enables SDXL-based models to transcend stylistic constraints. For instance, applying it to a model previously confined to anime generation now permits the creation of high-fidelity photorealistic images, among other capabilities. applying this to a anime only model allows it to generate high fiedelity photorealism and so on. |
|
|
|
The comprehensive rework of the latent representation has successfully addressed CFG scaling issues, allowing the model to effectively handle CFG ranges from 3 to 50. This eliminates the occurrence of total image failures, which were common at CFG levels above 10, without the need for additional steps. |
|
## Optimal Creative Settings |
|
|
|
CLIP Settings: |
|
Recommended settings include a CLIP of -2 and the strategic use of light negatives to unleash the full artistic potential of Proteus-RunDiffusion . |
|
|
|
CFG: |
|
For standard requests, a CFG setting of 8.5 is optimal, while artistic endeavors perform best at a CFG setting of 3.5. |
|
|
|
Tagging Versatility: |
|
The model supports the Pony tagging format and encourages experimentation. Users can explore tags like score_9, score_8_up, etc., or freely describe their creative vision. |
|
|
|
Backward Compatibility: |
|
Proteus-RunDiffusion’s adaptation of the Pony CLIP architecture guarantees backward compatibility with all SDXL models, offering a seamless and flexible creative experience. |
|
|
|
|
|
## Legal Disclaimer and Methodology |
|
|
|
Innovative Retraining Approach |
|
Proteus-RunDiffusionrepresents a cutting-edge development in AI-driven artistic creation. It is crucial to note that this model leverages a completely reworked and retrained version of Pony's CLIP model, achieved through a unique and undisclosed method developed specifically for this project. This retraining process is designed to enhance compatibility with all SDXL models, ensuring a broad and flexible application. |
|
|
|
Distinct from Pony's Unet |
|
For legal and ethical considerations, it is important to clarify that ProteusV0.4-RunDiffusion does not incorporate any Unet components originally developed by Pony. Our model’s innovative capabilities and improvements are the result of proprietary advancements and do not infringe upon Pony’s original Unet architecture. |
|
|
|
There will be a upcoming Human Preference Study and Research Publication |
|
- |
|
|
