{"post":{"slug":"stable-diffusion-3-5-guide-2026","title":"Stable Diffusion 3.5 guide for 2026: local generation, ComfyUI, and when to skip the cloud","excerpt":"SD 3.5 plus the ComfyUI ecosystem in 2026 produces output competitive with Flux Dev — locally, free of per-image fees. Here is the practical guide.","category":"ai","date":"February 1, 2026","author":"ByteLedger Team","content":"<span class=\"dropcap\">S</span>tability AI's SD 3.5 series shipped in late 2024 and matured through 2026 into a genuinely production-capable open-weights image stack — especially when paired with ComfyUI's node-based workflow engine. For teams generating thousands of images per month and willing to invest in local infrastructure, SD 3.5 + ComfyUI is the cheapest and most flexible option. Here is the practical guide.\n\n## What changed in 2026\n\n- **SD 3.5 Large + Turbo + Medium** cover the spectrum — 8B params for highest quality (Large), 2.5B for speed (Medium), 4-step Turbo for real-time UX.\n- **ComfyUI Manager** matured into the most-used SD interface, with thousands of community nodes.\n- **Permissive Stability community license** allows commercial use up to $1M annual revenue without paid license — covers most teams.\n\n## Hardware requirements\n\n| Model | VRAM | Speed (RTX 4090) |\n| --- | --- | --- |\n| SD 3.5 Medium | 8GB | ~2s/image |\n| SD 3.5 Large | 16GB | ~5-8s/image |\n| SD 3.5 Large + ControlNet | 20GB | ~10-15s/image |\n\nFor Mac users: M3 Pro / M4 Pro with 32GB+ unified memory runs SD 3.5 Large; M3 Max / M4 Max gives ~3x speed.\n\n## ComfyUI primer\n\nComfyUI is a node-graph interface for image generation pipelines. Each node does one thing (load model, encode prompt, sample, decode, save). Connect them; build complex pipelines.\n\nA basic SD 3.5 workflow has ~5 nodes:\n1. **Load Checkpoint** — pick SD 3.5 Large\n2. **CLIP Text Encode (positive + negative prompts)**\n3. **Empty Latent Image** — set dimensions\n4. **KSampler** — model + prompts + latent + steps + cfg\n5. **VAE Decode + Save Image**\n\nSave the workflow as JSON; reuse and modify.\n\n## Recommended ComfyUI setup\n\nInstall via Pinokio (one-click installer) or Comfy.org's official installer. Then install:\n- **ComfyUI-Manager** (essential for installing more nodes)\n- **rgthree-comfy** (better UX, multi-image preview)\n- **ComfyUI-AdvancedLivePortrait** for character consistency\n- **ControlNet Aux** for Canny, Depth, Pose preprocessing\n\n## SD 3.5 Large workflow patterns\n\n**Photorealistic**: cfg 4.5, steps 28, sampler dpmpp_2m. Negative prompt: `low quality, jpeg artifacts, watermark, low resolution`.\n\n**Stylized**: cfg 5.5, steps 35, sampler dpmpp_3m_sde. Add style LoRA at strength 0.8.\n\n**Fast iteration**: SD 3.5 Turbo, 4 steps, cfg 1.5, sampler dpm_sde_gpu. Real-time on RTX 4090.\n\n## LoRA training for brand-specific styles\n\nLoRAs (low-rank adaptations) are small (~50-200MB) trainable add-ons. Train one on 30-50 images of your brand style; produce on-brand output reliably. Tools: kohya-ss, OneTrainer, or ComfyUI's built-in trainer.\n\nTraining time: 1-2 hours on RTX 4090 for a quality LoRA.\n\n## SD 3.5 vs Flux vs Midjourney\n\n| Aspect | SD 3.5 Large | Flux Dev (local) | Midjourney v7 |\n| --- | --- | --- | --- |\n| Quality (1-10) | 7.5 | 8.5 | 9.0 |\n| Prompt adherence | 7.0 | 9.0 | 7.0 |\n| Text rendering | 6.0 | 8.5 | 5.0 |\n| Cost per 1000 images (local) | $0 | $0 | n/a |\n| Style flexibility | High (many LoRAs) | Medium | High |\n\nFor sheer image quality, Flux Dev usually wins. SD 3.5 wins on flexibility, ecosystem (10x more LoRAs and plugins), and zero per-image cost once you've set up local hardware.\n\n## When local makes sense\n\n**>500 images/mo:** local hardware pays back in 6-12 months vs API costs.\n\n**Sensitive content:** legal, medical, brand-internal imagery you don't want flowing through third-party APIs.\n\n**Heavy customization:** trained LoRAs, custom ControlNet variants, complex workflows. Cloud APIs typically don't expose these.\n\n**Skip local if:** you generate <200 images/mo. API services like Replicate, fal, or Together cost $5-25/mo for that volume.\n\n## FAQ\n\n**Can I use SD 3.5 commercially?**\nYes if your annual revenue is under $1M. Above that, Stability requires a license (~$2K/mo for $1M-$10M revenue).\n\n**What about ControlNet for SD 3.5?**\nControlNets for SD 3.5 are available for Canny, Depth, Pose, and Tile via Stability and community releases.\n\n**Is SDXL still relevant?**\nLess so in 2026. SD 3.5 Large beats SDXL on most metrics. SDXL still has a richer LoRA ecosystem (more legacy options).\n\n## Where to go next\n\nFor related guides see [Midjourney v7 prompting guide](/blog/midjourney-v7-prompting-guide-2026), [Flux Pro prompting guide](/blog/flux-pro-prompting-guide-2026), and [Local LLM setup guide for 2026](/blog/local-llm-setup-guide-2026).","hue":218,"metaTitle":"Stable Diffusion 3.5 guide (2026): local & ComfyUI","metaDescription":"Practical 2026 guide to Stable Diffusion 3.5 — hardware, ComfyUI workflows, LoRA training, and when local generation beats API services.","keywords":["stable diffusion 3.5","sd 3.5 guide","comfyui 2026","stable diffusion local","sd 3.5 hardware","sd 3.5 vs flux","stable diffusion lora","free image generation 2026","stability ai 2026","open source image gen"],"tldr":["**SD 3.5 Large is the strongest open SD model** with permissive commercial use under the Stability community license.","**ComfyUI in 2026** is the de-facto interface for serious workflows — node-based, scriptable, and powerful.","**Hardware: 16GB VRAM minimum for SD 3.5 Large.** RTX 4090, RTX 5090, or M-series Macs with 32GB+ unified memory.","**Skip:** SD 3.5 if your workload is small (<200 images/mo) — API services are cheaper than the time investment."],"readTime":"3 min read","wordCount":698},"related":[{"slug":"how-to-use-stable-diffusion-2026","title":"How to use Stable Diffusion in 2026: setup, prompting, and real results","excerpt":"Stable Diffusion remains the most customizable open-source image model in 2026. Here is how to install it, prompt effectively, and get professional-quality results fast.","category":"ai","date":"February 28, 2026","author":"ByteLedger Team","readTime":"4 min read","hue":206},{"slug":"stable-diffusion-vs-midjourney-2026","title":"Stable Diffusion vs Midjourney in 2026: control or polish?","excerpt":"Stable Diffusion gives you open, local control; Midjourney gives you the best default look. Here is which AI image tool fits your work in 2026, and why.","category":"ai","date":"June 1, 2026","author":"ByteLedger Team","readTime":"3 min read","hue":290},{"slug":"what-is-stable-diffusion-2026","title":"What Is Stable Diffusion? Explained Plainly for 2026","excerpt":"Stable Diffusion is the open-weights image model you can run yourself. Here is what it is, how it works, and when it beats a paid tool in 2026.","category":"ai","date":"April 30, 2026","author":"ByteLedger Team","readTime":"4 min read","hue":280}],"sameCategory":[{"slug":"agent-cost-attribution-2026","title":"Agent Cost Attribution: Finding Where the Tokens Went","excerpt":"A single monthly API bill tells you that agents are expensive. It does not tell you which feature, which customer, or which retry loop is responsible. That gap is fixable.","category":"ai","date":"August 19, 2026","author":"ByteLedger Team","readTime":"5 min read","hue":220},{"slug":"human-in-the-loop-agents-2026","title":"Human-in-the-Loop Agents: Where to Put the Approval Gate","excerpt":"Approve everything and the agent is a slow autocomplete. Approve nothing and you have handed irreversible actions to a system that can be talked into things.","category":"ai","date":"August 19, 2026","author":"ByteLedger Team","readTime":"5 min read","hue":197},{"slug":"reasoning-effort-controls-2026","title":"Reasoning Effort Controls: Tuning How Hard a Model Thinks","excerpt":"Fixed thinking budgets gave way to effort levels, and the default is rarely the right setting. Here is how the dial behaves and where each level actually belongs.","category":"ai","date":"August 19, 2026","author":"ByteLedger Team","readTime":"5 min read","hue":210},{"slug":"retrieval-metrics-explained-2026","title":"Retrieval Metrics: Measuring the Half of RAG That Fails","excerpt":"When a RAG system answers badly, the retrieval usually failed before the model ever spoke. Measuring generation quality alone hides that entirely.","category":"ai","date":"August 19, 2026","author":"ByteLedger Team","readTime":"5 min read","hue":215}],"newer":{"slug":"series-i-bonds-guide-2026","title":"I Bonds in 2026: are they still worth buying?","excerpt":"I Bonds had a moment in 2022 with 9.62% rates. In 2026, the composite rate has settled considerably lower. Here is whether they still make sense, who they make sense for, and the buying mechanics.","category":"finance","date":"February 1, 2026","author":"ByteLedger Team","readTime":"4 min read","hue":282},"older":{"slug":"starlink-v3-review-2026","title":"Starlink in 2026: is it finally good enough to replace home internet?","excerpt":"Starlink V3 satellites and the new flat high-performance dish have materially improved performance since 2024. For many rural and semi-rural users in 2026, the question is no longer \"is it good enough\" but \"is anything else available.\"","category":"tech","date":"February 1, 2026","author":"ByteLedger Team","readTime":"4 min read","hue":232}}