Stable Diffusion 3.5 Large is Stability AI's 8-billion-parameter open image model, free under the Stability Community License for anyone or any organisation under $1M in annual revenue (an Enterprise License is required above that). ComfyUI is the free, node-based interface commonly used to run it. Pick it when you want self-hosted image generation with quantization options for lower-VRAM GPUs, and are willing to learn a visual node graph instead of a simple prompt box.
Other good fits
Building a repeatable, inspectable image-generation workflow visually, rather than a single black-box prompt.
+ Pros
Read pros
Stable Diffusion 3.5 Large is an 8-billion-parameter, open-weight text-to-image model from Stability AI, built as a Multimodal Diffusion Transformer with three fixed text encoders. It's licensed under the Stability Community License: free for research, non-commercial use, and commercial use by individuals or organisations with less than $1M in total annual revenue; above that threshold, an Enterprise License from Stability is required. ComfyUI is the free, open-source, node-based interface most commonly used to run it, letting you visually wire together the model, prompts, samplers and post-processing steps into a repeatable graph.
Pick it when…
You want to self-host image generation and can work within the Community License's revenue threshold.
You want to build a repeatable, inspectable generation workflow (a saved node graph) rather than a one-off prompt.
Your GPU has limited VRAM; the model supports 4-bit quantization and CPU offloading to fit smaller hardware.
− Cons
Read cons
ComfyUI's node-graph interface has a genuine learning curve compared to a single prompt box; expect to spend real time understanding node types and connections before you're productive, especially for anything beyond the most basic text-to-image flow. The Community License's $1M revenue threshold is a real legal boundary, not a suggestion: check your organisation's actual revenue against it rather than assuming a small team automatically qualifies as under the threshold indefinitely.
Pick something else when…
Your organisation exceeds the $1M revenue threshold. You need Stability's Enterprise License, not the free Community terms.
You want a simpler interface without learning a node graph. A straightforward prompt-box tool skips ComfyUI's learning curve, at the cost of less workflow flexibility.
You want a different open image model to compare quality and license terms against.Flux-2 is a comparable open-weight alternative worth checking.
Avoid if
You want a simple prompt box; ComfyUI's node graph has a real learning curve compared to a one-field interface.
Workflow
Install ComfyUI and download the SD 3.5 Large checkpoint → build a basic graph (checkpoint loader → prompt → sampler → save image) → generate a test image → adjust sampler settings and prompt based on the result → save the working graph as a reusable template for similar future generations.
Cost
The model weights and ComfyUI are both free under their respective licenses (Stability Community License for the model, open source for ComfyUI). You pay for the machine: SD 3.5 Large supports 4-bit quantization via BitsAndBytes and CPU offloading to reduce VRAM needs, so it can run on more modest GPUs than a full-precision 8B model would otherwise need, at some speed and possibly quality cost. If your organisation exceeds $1M in annual revenue, budget for Stability's Enterprise License instead of assuming the free Community terms apply.
Restrictions
Many published ComfyUI workflows depend on separately installed custom nodes that aren't part of the base install; node compatibility can break across ComfyUI or model updates, so pin versions for anything you need to keep working.
First project
Documentation-based suggestion — not run here.
Build and save a basic text-to-image workflow.
Install ComfyUI and download the SD 3.5 Large model weights.
Build the minimal graph: load checkpoint, encode a text prompt, sample, save image.
Generate one image from a specific, detailed prompt (not a vague one).
Save the graph as a template, then generate a second image by only changing the prompt text, confirming the workflow is actually reusable.
Done when you have a saved, reusable ComfyUI graph that produces a new image just by swapping the prompt, without rebuilding the node connections each time.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on the Stable Diffusion 3.5 Large model card and the ComfyUI repository, checked 23 September 2026. No installation, workflow build or image generation was performed for this review.
Stable Diffusion 3.5 Large model card · Sep 23, 2026 — Stability Community License ($1M revenue threshold, Enterprise License above), 8B parameters, MMDiT architecture, 4-bit quantization and CPU offloading.
Build repeatable image workflows by connecting generation, ControlNet, LoRA, and upscaling nodes.
Run Stable Diffusion 3.5 locally through ComfyUI's saved visual graphs.
Choose model sizes that match your available consumer or larger GPU.
Stability AI (models); Comfy Org / comfyanonymous (the ComfyUI interface) · free · self-host
node-graph workflow
offline generation
image upscaler
controlnet integration
lora fine-tuning
gpu inference
free commercial use
open-source
draft-docs
View sourced imageOfficial ComfyUI workflow screenshot from its Stable Diffusion 3.5 Medium announcement, 29 October 2024. The publisher's example nodes and output are retained.
Stable Diffusion 3.5 + ComfyUI official product documentation
Owner-requested editorial use of public official interface screenshots, 15 September 2026. Copyright retained by the original publisher; attribution preserved. No open licence or hands-on course execution is asserted. Source asset: https://substack-post-media.s3.amazonaws.com/public/images/66bba561-e5a9-4a59-a735-31fc54515823_2048x918.png. PNG conversion preserves the complete frame. Converted on 15 October 2026 to a 1600px-wide WebP (quality 80) for faster page loads; the original image remains in the media library.