TensorArt-TurboX Series Release Announcement 🎉
Hello everyone, we are very excited to announce that we have just open-sourced SD3.5 TurboX! This update highlights the release of two efficient models, designed to bring the community a faster and higher-quality image generation experience. 🚀
TurboX Limited-Time Event 🎊
To help us test the performance of Turbo X and build our ecosystem, we are launching a limited-time event.
- Generate images using Turbo X and post:
- Earn 50 Credits.
- Publish an AI Tool using Turbo X:
- Receive an additional 200 Credits.
Participation Requirements: When posting your images and AI Tools, please add the tag TURBOX.
Event Period: March 6th to March 12th.
Rewards Distribution: Rewards will be distributed within 7 days after the event concludes on March 12th.
Overview
The TensorArt-TurboX Series includes two models:
-
SD3.5 Large TurboX: Uses 8 sampling steps to deliver a 6x speed boost over the original model⚡, while achieving superior image quality compared to the official Stable Diffusion 3.5 Turbo. Huggingface
-
SD3.5 Medium TurboX: With just 4 sampling steps, this model generates 768x1248 resolution images in 1 second on mid-range GPUs (e.g., RTX3080), realizing a 13x speed improvement over the original. Huggingface
Multiple Versions Available
The SD3.5 Large TurboX model is offered in both LoRA and ckpt versions. It has been tested for compatibility with most community models, facilitating smoother integration and faster prototyping across diverse projects.
Enhanced Visual Quality
SD3.5 Large TurboX stands out in image diversity, richness, and realism—outperforming the official Stable Diffusion 3.5 Turbo in human detail enhancement.
1. SD3.5 Large TurboX 🌟
Usage Instructions
- Model Selection: Choose the LoRA version
Tensorart-Turbo-SD3.5Largewith a strength of 1. - Sampler:
euler - Scheduler:
simple - Sampling Steps:
8 steps - CFG Scale:
Between 1 and 1.5(helps enhance details, particularly in human hands) - LoRA Strength: 1.0
Model Features
-
Speed: Achieves a 6x faster generation speed compared to the original SD3.5 Large, with minimal quality loss.
Note: When CFG ≠ 1, generation speed can double compared to CFG=1. -
Superior Quality: Outperforms the official Stable Diffusion 3.5 Turbo in image detail, diversity, richness, and realism.
-
Versatility: Available in both ckpt and LoRA formats, making it easy to integrate with most realistic and anime-style models in the community, thereby accelerating workflows and tool development.
Not Recommended For
- Scenarios that require precise rendering of English text.
- Tasks demanding flawless human hand details.
- Users experimenting with various samplers or noise schedulers, since the model's distilled performance is based on the specific configuration (Euler simple with shift=5).
Additional Note:
SD3.5 Large TurboX also demonstrates exceptional performance in image diversity, richness, and realism, outperforming flux-dev in human detail enhancement.
2. SD3.5 Medium TurboX
Highlights
- 4 Sampling Steps: The
tensorart_sd3.5m_4stepsversion achieves the quality of 25+ steps with CFG=1, but in just 4 steps. - Unmatched Speed: Generates a 768x1248 image in only 1 second on mid-range GPUs like the RTX3080—a 13x speed improvement over the original model.
Usage
- Follow the example settings provided in the reference image for optimal results. 👍
Resources
-
SD3.5 Large TurboX:
-
SD3.5 Medium TurboX:
-
Workflow:
Conclusion
We invite developers and artists alike to try out the new TensorArt-TurboX series and share your feedback.
Let's work together to push the boundaries of open-source AI art generation! 🤝
TensorArt-TurboX Series Release Announcement 🎉
Hello everyone, we are very excited to announce that we have just open-sourced SD3.5 TurboX! This update highlights the release of two efficient models, designed to bring the community a faster and higher-quality image generation experience. 🚀
TurboX Limited-Time Event 🎊
To help us test the performance of Turbo X and build our ecosystem, we are launching a limited-time event.
- Generate images using Turbo X and post:
- Earn 50 Credits.
- Publish an AI Tool using Turbo X:
- Receive an additional 200 Credits.
Participation Requirements: When posting your images and AI Tools, please add the tag TURBOX.
Event Period: March 6th to March 12th.
Rewards Distribution: Rewards will be distributed within 7 days after the event concludes on March 12th.
Overview
The TensorArt-TurboX Series includes two models:
-
SD3.5 Large TurboX: Uses 8 sampling steps to deliver a 6x speed boost over the original model⚡, while achieving superior image quality compared to the official Stable Diffusion 3.5 Turbo. Huggingface
-
SD3.5 Medium TurboX: With just 4 sampling steps, this model generates 768x1248 resolution images in 1 second on mid-range GPUs (e.g., RTX3080), realizing a 13x speed improvement over the original. Huggingface
Multiple Versions Available
The SD3.5 Large TurboX model is offered in both LoRA and ckpt versions. It has been tested for compatibility with most community models, facilitating smoother integration and faster prototyping across diverse projects.
Enhanced Visual Quality
SD3.5 Large TurboX stands out in image diversity, richness, and realism—outperforming the official Stable Diffusion 3.5 Turbo in human detail enhancement.
1. SD3.5 Large TurboX 🌟
Usage Instructions
- Model Selection: Choose the LoRA version
Tensorart-Turbo-SD3.5Largewith a strength of 1. - Sampler:
euler - Scheduler:
simple - Sampling Steps:
8 steps - CFG Scale:
Between 1 and 1.5(helps enhance details, particularly in human hands) - LoRA Strength: 1.0
Model Features
-
Speed: Achieves a 6x faster generation speed compared to the original SD3.5 Large, with minimal quality loss.
Note: When CFG ≠ 1, generation speed can double compared to CFG=1. -
Superior Quality: Outperforms the official Stable Diffusion 3.5 Turbo in image detail, diversity, richness, and realism.
-
Versatility: Available in both ckpt and LoRA formats, making it easy to integrate with most realistic and anime-style models in the community, thereby accelerating workflows and tool development.
Not Recommended For
- Scenarios that require precise rendering of English text.
- Tasks demanding flawless human hand details.
- Users experimenting with various samplers or noise schedulers, since the model's distilled performance is based on the specific configuration (Euler simple with shift=5).
Additional Note:
SD3.5 Large TurboX also demonstrates exceptional performance in image diversity, richness, and realism, outperforming flux-dev in human detail enhancement.
2. SD3.5 Medium TurboX
Highlights
- 4 Sampling Steps: The
tensorart_sd3.5m_4stepsversion achieves the quality of 25+ steps with CFG=1, but in just 4 steps. - Unmatched Speed: Generates a 768x1248 image in only 1 second on mid-range GPUs like the RTX3080—a 13x speed improvement over the original model.
Usage
- Follow the example settings provided in the reference image for optimal results. 👍
Resources
-
SD3.5 Large TurboX:
-
SD3.5 Medium TurboX:
-
Workflow:
Conclusion
We invite developers and artists alike to try out the new TensorArt-TurboX series and share your feedback.
Let's work together to push the boundaries of open-source AI art generation! 🤝