Storyboard Generation
Sequential image generation for film storyboards, comic panels. Consistency across frames is the hard problem.
Setup walkthrough
- Install ComfyUI via Stability Matrix.
- ComfyUI Manager → Install Models → "flux1-dev" (23 GB) or "sdxl-base" (7 GB) for lighter weight.
- Storyboard workflow: you need consistent characters across frames. Two approaches:
Simple: Use a fixed seed + fixed prompt structure. Prompt: "[SHOT 1] Wide shot of detective in office, raining outside, moody lighting" → generate. Change prompt to "[SHOT 2] Close-up of detective's hand on evidence" → same seed.
Advanced: Use IP-Adapter (install via ComfyUI Manager) with a reference image of your character. Each frame: same IP-Adapter → consistent face/clothing.
- First storyboard frame in 8-15 seconds (SDXL) or 10-20 seconds (Flux) on 12+ GB GPU. A 6-panel storyboard in 1-3 minutes.
- For text annotations on frames: add "Scene 1, INT. OFFICE - DAY" via the prompt itself or overlay in post.
The cheap setup
Used RTX 3060 12 GB ($200-250, see /hardware/rtx-3060-12gb). Runs SDXL for storyboard frames at 8-15 seconds each — a 12-panel storyboard in ~2-3 minutes. IP-Adapter for character consistency adds minimal overhead (uses <1 GB VRAM). Pair with Ryzen 5 5600 + 32 GB DDR4 + 1TB NVMe. Total: ~$390-440. For simple storyboards (stick figure level, layout exploration), even a GTX 1060 6 GB ($60) with SD 1.5 does 3-6 seconds per frame — useful for blocking and composition planning.
The serious setup
Used RTX 3090 24 GB ($700-900, see /hardware/rtx-3090). Runs Flux Dev for high-quality storyboard frames at 10-20 seconds each. Can load multiple LoRAs (character + environment + style) simultaneously for frame-to-frame consistency. For film production storyboarding (50-100 frames/day): the workflow is viable. Total: ~$1,800-2,200. For the absolute fastest iteration: RTX 4090 ($2,000) at 5-8 seconds per Flux frame. Storyboard generation is a throughput problem — more frames = more ideas explored.
Common beginner mistake
The mistake: Generating each storyboard frame with a new random seed, getting a different "actor" and lighting style in every panel. Why it fails: Random seeds are random — each frame is a completely independent generation with different facial features, clothing, and lighting. The storyboard looks like scenes from 6 different movies glued together. The fix: Lock the seed across frames. Use {$seed:12345} in ComfyUI for all panels. The same seed + same model produces consistent style. Then vary the prompt for content while the seed maintains visual coherence. For character consistency: use IP-Adapter with a reference image. For environment consistency: use ControlNet depth/canny with a 3D blockout render as input for each frame.
Recommended setup for storyboard generation
Browse all tools for runtimes that fit this workload.
Reality check
Image gen is compute-bound, not bandwidth-bound. VRAM matters for the resolution + LoRA training stack, but FP16 TFLOPS is what decides Flux throughput. The 5080's compute advantage over 5070 Ti shows here in ways it doesn't on LLM inference.
Common mistakes
- Buying for VRAM ceiling without checking compute (16 GB Flux Dev FP16 doesn't fit anyway)
- Skipping LoRA training requirements (24 GB minimum, 32 GB comfortable for Flux)
- Underestimating ComfyUI's multi-model VRAM appetite vs A1111's single-pipeline
- Using Q4 quantized image models — quality drop is more visible than on LLMs
What breaks first
The errors most operators hit when running storyboard generation locally. Each links to a diagnose+fix walkthrough.
Before you buy
Verify your specific hardware can handle storyboard generation before committing money.