Model Card for Recursed Canvas
Model Details
Model Description
Recursed Canvas is a Stable Diffusion 1.5 model primarily tuned for artistic depictions of women's fashion, beauty, and fantasy art. Developed by the user odyss3y, this model excels at generating artistic (e.g., digital illustrations, 3d renders) images, particularly within the uncanny valley of realism. The intention behind Recursed Canvas is not to achieve photo-quality images but to push the boundaries of realism within the context of AI-generated art. This model is designed to be more realistic than its counterpart, Recursed Canvas, with which it is frequently blended.
Blending Recursed Canvas with Aberrated Perceptron using tools like A1111’s refiner, hires, and in-painting techniques (including Adetailer) has yielded fascinating results. The blending process typically involves using Recursed Canvas as the base for the first 20-95% of the image generation, with Aberrated Perceptron adding its unique touch of photorealism.
The model serves as a benchmark in exploring the strengths, weaknesses, and limitations of realistic image generation within the Stable Diffusion 1.5 framework.
🖼 Sample images from this model can be found on
Civitai.com 🖼
- Developed by: odyss3y
- Funded by: Personal
- Shared by: odyss3y
- Model type: Stable Diffusion 1.5, focused on fashion, beauty, and fantasy art
- Language(s) (NLP): N/A
- License: CreativeML OpenRAIL-M
- Finetuned from model: N/A (Based entirely on merges with other models)
Model Sources
- Repository: Refer to the changelog for specific version information.
Uses
Direct Use
Recursed Canvas is intended for generating art with a female focus, emphasizing beauty, fashion, and fantasy storytelling. It is ideal for artists and creatives seeking to explore hyper-realistic yet artistically enhanced imagery within these domains.
Out-of-Scope Use
The model should not be used in ways that violate ethical guidelines, including but not limited to the creation of misleading or harmful content, especially involving real people or depictions of children. Users must comply with all relevant laws and ethical standards when using this model.
Bias, Risks, and Limitations
The model has been overfitted with imagery of beautiful women, likely resulting in biased outputs that skew towards this direction. Users should be aware of this bias when generating images and consider the ethical implications of such biases.
Recommendations
Users should carefully consider the potential biases and ethical concerns when using Recursed Canvas. It is recommended to avoid generating content that could be harmful or offensive, particularly involving sensitive subjects like real individuals or minors.
How to Get Started with the Model
Use the following code to get started with Recursed Canvas:
1# Example code to load the model
2from diffusers import StableDiffusionPipeline
3
4model_id = "path_to_aberrated_perceptron_model"
5pipe = StableDiffusionPipeline.from_pretrained(model_id)
6pipe.to("cuda")
7
8prompt = "a hyper-realistic portrait of a woman in a fantasy setting"
9image = pipe(prompt).images[0]
10image.show()
Training Details
Training Data
This model is based entirely on merges with other models, leveraging their strengths to achieve the desired effects in image generation. There was no specific dataset used for fine-tuning or additional training.
Training Procedure
No direct training was performed on Recursed Canvas. The model's capabilities were shaped through the careful merging of existing models, using techniques such as Merge Block Weighting (MBW).
Preprocessing
N/A
Training Hyperparameters
- Training regime: N/A (No direct training performed)
Speeds, Sizes, Times
- Merging Time: Momentary, depending on the merge method used.
- Evaluation Time: Approximately 18 hours of GPU time on an NVIDIA GeForce RTX 3060 Ti.
Evaluation
Testing Data, Factors & Metrics
Testing Data
Standardized prompts were used to evaluate the model's performance, particularly focusing on the effects of various merge processes.
Factors
The evaluation considered the realism and artistic quality of generated images, disaggregated by variations in merge processes and prompt adjustments.
Metrics
The primary evaluation metrics were subjective assessments of image realism, alignment with artistic goals, and the ability to generate desired effects within the model's intended scope.
Results
The model successfully achieved its intended purpose of generating hyper-realistic images within the uncanny valley of realism, with an emphasis on fashion, beauty, and fantasy art. The merging process with Recursed Canvas provided a versatile toolset for balancing realism and artistic expression.
Environmental Impact
The carbon emissions for training and evaluating Recursed Canvas can be estimated using the
Machine Learning Impact calculator.
- Hardware Type: NVIDIA GeForce RTX 3060 Ti
- Hours used: Approximately 18 hours
- Cloud Provider: N/A (Local workstation in California)
- Compute Region: California, USA
- Carbon Emitted: [Calculation Needed]
Technical Specifications
Model Architecture and Objective
Recursed Canvas utilizes the Stable Diffusion 1.5 architecture with a focus on realistic yet artistically enhanced image generation in the domains of fashion, beauty, and fantasy.
Compute Infrastructure
Hardware
- NVIDIA GeForce RTX 3060 Ti
Software
- A1111 with refiner, hires, and in-painting techniques (Adetailer)
Citation
BibTeX:
[More Information Needed]
APA:
[More Information Needed]
Glossary [optional]
N/A
More Information [optional]
[More Information Needed]
Model Card Authors
Model card authored by odyss3y.
Model Card Contact
For inquiries, please contact [contact information if desired].
