Stable Diffusion 3.5 Medium
Stability AI's 2.5B-parameter SD3.5 with strong quality/speed trade-off. Consumer-GPU friendly.
Stable Diffusion 3.5 Medium is image generation AI model from Custom, priced at β¬0.000 per 1M input tokens with a unknown context window.
Pricing
API Integration
Use our OpenAI-compatible API to integrate Stable Diffusion 3.5 Medium into your application.
npm install railwailimport railwail from "railwail";
const rw = railwail("YOUR_API_KEY");
const images = await rw.run("sd-3-5-medium-stability", "A beautiful sunset over Tokyo");
console.log(images[0].url);
// Or use the image() method for full control
const res = await rw.image("sd-3-5-medium-stability", "A cat in space", {
size: "1024x1024",
n: 1,
});
console.log(res.data[0].url);Deep dive β Stability AI's Stable Diffusion 3.5 Medium
Stability AI was founded in 2019 by Emad Mostaque (CEO until March 2024; succeeded by Prem Akkaraju) and is headquartered in London. The company sponsored and released the original Stable Diffusion (2022, with CompVis/LMU Munich and Runway), SDXL (2023), SD3 (2024) and SD 3.5 (Oct 2024). The October 2024 SD 3.5 release shipped three variants β Large (8.1B), Large Turbo (distilled few-step) and Medium (2.5B) β all under a permissive Community License.
Visit Stability AI βStable Diffusion 3.5 Medium is the mid-tier model of the October 2024 SD 3.5 release. It uses a refined architecture Stability AI calls MMDiT-X, a 2.5B-parameter Multimodal Diffusion Transformer with improved positional embeddings and joint self-attention designed to fix the anatomical issues that plagued the earlier SD3 Medium 2B release. The model is trained with the rectified-flow / flow-matching objective described in the SD3 paper (Esser et al. 2024) and uses three parallel text encoders β CLIP-L, CLIP-G and T5-XXL β for strong long-prompt understanding. SD 3.5 Medium targets a sweet spot of quality and inference cost: it runs comfortably on a single 12GB consumer GPU, supports 1024x1024 native resolution, and is fast enough for interactive use. It is also intended to be the most fine-tune-friendly member of the SD 3.5 family, with a smaller training footprint than the Large variant.
- Parameters
- 2.5B parameters
- Context
- 256 tokens
- Runs on consumer GPUs (12GB VRAM with FP16)
- Open weights under Stability AI Community License
- MMDiT-X architecture fixes SD3 anatomy issues
- Three text encoders for strong prompt adherence
- 1024x1024 native resolution
- Fine-tune-friendly thanks to smaller footprint
- Compatible with ControlNet, LoRA, IP-Adapter ecosystem
- Best for: fine-tuning base, consumer apps, on-device generation, education, prototyping.
Pretrained on a large licensed and publicly available image-text dataset; filtered for safety and quality. Smaller training footprint than SD 3.5 Large.
License: Stability AI Community License β free for non-commercial and commercial use up to $1M annual revenue; enterprise license required above that threshold.
Known limitations
- Lower quality than SD 3.5 Large on photorealism
- Smaller capacity limits very complex compositions
- Open weights have no built-in safety filter
- Commercial-revenue threshold ($1M) for free use
Frequently asked questions
Related Models
View all Image GenerationFLUX 1.1 Pro
Black Forest Labs' flagship text-to-image model. Faster generation than FLUX.1 Pro at higher prompt adherence, with strong photorealism and reliable spatial composition. Runs as a hosted Replicate model.
FLUX 1.1 Pro Ultra
FLUX 1.1 Pro in Ultra mode by Black Forest Labs. Generates up to 4 megapixel images with a raw mode for less processed, more natural-looking photography. Best FLUX option when output resolution and fine detail matter.
Flux 1.1 Pro Ultra
FLUX 1.1 Pro in ultra mode. Up to 4 megapixel images with raw mode for photorealism.
Flux Dev
Black Forest Labs' development model. Fast, high-quality image generation with LoRA support.
Start using Stable Diffusion 3.5 Medium today
Get started with free credits. No credit card required. Access Stable Diffusion 3.5 Medium and 100+ other models through a single API.