The Claude + Seedance UGC workflow splits production into two roles: Claude directs (brief, shot list, prompts, narrative review) and Seedance generates (5-second photoreal clips at up to 1080p). The handoff is structured prompts; the output is platform-ready ad creative.
Photoreal AI UGC ads work on Meta, TikTok, and YouTube because the format mirrors how real creators shoot: handheld phone, natural lighting, one cut every 5 seconds, caption overlay. The barrier was always production: sourcing creators, briefing, reshooting, paying per asset. AI removes that barrier, but only when the prompts are tight enough to produce something that does not look like an AI ad.
Claude is unusually good at writing those prompts. Given a brand voice file and an ICP, it produces a 3-shot structure (hook, demo, payoff), the right camera language, and the lighting and setting anchors that lock the look across the spot. Seedance is unusually good at executing those prompts: 1080p output, multi-shot storytelling, smooth motion, and a Pro tier that handles cinematic detail when needed. The combination produces UGC that scrolls past Meta’s synthetic-media detection bias when the prompt anchors are right.
This guide is the step-by-step process. The 7 steps below produce a 15-to-30 second spot in 30 to 60 minutes the first time, and 15 to 30 minutes for variations. The six prompt patterns below are the anchors that hold brand and character continuity across shots.
What Seedance Actually Does
ByteDance describes Seedance as a model that "supports multi-shot video generation from both text and image" with breakthroughs in "semantic understanding and prompt following" producing "1080p videos with smooth motion, rich details, and cinematic aesthetics." The headline capabilities are multi-shot storytelling, photorealism plus stylised modes (cyberpunk, illustration, felt texture), multi-agent interactions, and a wide camera-movement vocabulary.
The two product tiers that matter for UGC: Seedance 1.0 Pro outputs 5-second 1080p clips at roughly $0.62 per video on the Fal.ai host, with image-to-video and text-to-video both supported. Seedance 2.0 ships ByteDance's most advanced text-to-video and image-to-video models with native audio, multi-shot editing, real-world physics, and director-level camera control. For most UGC ad workflows, Seedance 1.0 Pro at 5-second 1080p is enough; reach for 2.0 when you need native audio, physics fidelity, or a single multi-shot generation instead of stitching.
5-second clips at 1080p (Pro) or 720p (Lite). Multi-shot storytelling on 2.0.
Text-to-video, image-to-video. Reference-to-video on 2.0 with up to 9 images, 3 videos, 3 audio clips.
Photoreal plus cyberpunk, illustration, felt texture, and other stylised modes.
Why Claude Is the Right Director
A video model produces what it is told. The bottleneck on AI UGC has never been the model; it has been the prompt. Generic prompts produce generic faces and generic settings. The whole craft of AI ad creative is writing prompts that lock subject, camera, lighting, setting, and style across multiple shots so the output reads as one ad and not three unrelated clips.
Claude is unusually strong at structured prompt writing because the model holds the brief, the audience, and the brand voice in context across the session. A single conversation produces the 3-shot structure, the per-shot prompts, the caption copy, and the CTA, all anchored against the same brief. The operator does not have to context-switch between "writing the script" and "writing the prompt" — they are the same step.
The 7-Step Workflow
Open Claude with the brand voice file loaded
Start a Claude session with your brand-voice and ICP files in context. Either paste them into the chat, attach as project knowledge in Claude.ai, or load via a SKILL.md if you run on Claude Code. Claude needs the voice anchor and the audience anchor before it can write a prompt that does not sound generic.
Brief Claude on the ad goal in three lines
State the product, the audience pain, and the conversion ask. Example: "Product is a $39/mo email-warmup tool. Audience is solo founders cold-emailing prospects. Pain is replies going to spam. Goal is a 20-second UGC ad ending with sign-up."
Ask Claude for a 3-shot UGC structure (hook, demo, payoff)
Hook is the 0-3 second attention pattern (a problem statement, a counterintuitive claim, a "wait, this works?" reaction). Demo is the 3-15 second product moment (the screen, the product in hand, the result on the screen). Payoff is the 15-20 second close (one-line resolution, on-screen offer, CTA). Claude returns this as a structured table with timing per shot.
Have Claude write a Seedance prompt for each shot
For each of the 3 shots, ask Claude to write a Seedance prompt that names the subject, action, camera framing, lighting, setting, and style modifier. Example output: "Close-up handheld phone selfie of a 28-year-old woman in a sunlit kitchen, holding her laptop, looking frustrated at the screen, soft natural daylight from a window on the right, photoreal UGC iPhone footage aesthetic, no cinematic gloss."
Generate each shot in Seedance
Paste each prompt into a Seedance-enabled platform. The ByteDance native interface, the Fal.ai Seedance hosts, or any other partner that exposes the model. Generate at 1080p for the hero shot and 720p for B-roll if you are budget-tuning. Iterate the prompt 1 to 2 times if the first generation drifts from the brief. Save the seed value of the best take so the same character anchors carry to subsequent shots.
Stitch, caption, and add audio
Drop the 3 clips into any standard editor (CapCut, Descript, Adobe Express, or an AI video workspace). Add burnt-in captions for the hook and CTA. Layer one audio track: either Claude-written voiceover from a TTS model or licensed background music depending on whether the spot uses a voice-over creative or a non-talking-head photoreal style. Export 1080x1920 for Reels, TikTok, and Shorts.
Tag the export for attribution before publishing
Inject a UTM string in the on-screen CTA URL or a hidden experiment ID in the asset filename. Without attribution, "the ad worked" collapses into vibes. With attribution, the 30-day audit can rank which Claude-prompted variants converted and prune the rest. This step is the difference between a workflow that compounds and a workflow that stays a hobby.
The 30-minute target assumes the brand voice and ICP files are already written. First-time setup of those anchors takes another 30 to 60 minutes once and amortises across every ad after.
The 6 Prompt Anchors That Hold Continuity
Six anchors form the prompt template Claude writes against. Every Seedance generation in the same ad set should reuse the same anchors with only the action changed. The template lives in a SKILL.md or a project knowledge file so it survives sessions.
Subject anchor
Name age, ethnicity if relevant to the audience, body type, clothing style, and one personality cue. "A 28-year-old woman in casual hoodie, slight grin, looks like a typical scroll-through Instagram user, not a model." Generic descriptions return generic faces. Specific anchors return characters that fit the ad.
Camera anchor (UGC vs cinematic)
For UGC: handheld, phone selfie angle, slightly shaky, vertical framing. For cinematic: tripod-locked, low or medium-angle establishing shot, rule-of-thirds composition. Mixing these two collapses the UGC effect. Pick one and apply it across the shot list.
Lighting anchor
Soft natural daylight from a window. Golden hour outdoor sun. Warm tungsten indoor evening. Cold morning overcast. Each lighting setup signals a different time-of-day and emotional tone. Lock one across the ad set so the three shots feel like one moment.
Setting anchor
Specific real-world setting beats vague "modern interior." A sunlit kitchen. A bus-stop sidewalk in October. A coffee shop window seat. Real settings lock the audience into a scene; abstract spaces look stock and AI-generated.
Style modifier
For Seedance UGC: "photoreal UGC iPhone footage aesthetic, no cinematic gloss, no Hollywood lighting, slight motion blur, natural skin texture." Negative-style modifiers cut Seedance's default tendency toward overly polished output, which kills the UGC effect.
Action anchor (single verb per shot)
One clear verb per 5-second shot. Reading the screen. Holding the product. Smiling at the camera. Pouring the coffee. Multiple actions in one prompt produce blurred motion. Lock one action per shot and let the editor stitch them into a sequence.
Worked Example: 20-Second Email-Warmup Ad
A worked example shows the structure end-to-end. Brief: $39/mo email-warmup tool, audience is solo founders cold-emailing, pain is replies going to spam, goal is a 20-second UGC ad ending with sign-up. Claude produces the following.
Shot 1 (0-5s) — Hook
Close-up handheld phone selfie of a 32-year-old woman in a hoodie, sitting at a kitchen table, looking visibly annoyed at her laptop screen, soft natural daylight from a window on the right, photoreal UGC iPhone footage aesthetic, no cinematic gloss.
Caption overlay: "I sent 200 cold emails. Got 0 replies."
Shot 2 (5-15s) — Demo
Same 32-year-old woman, same kitchen table, now leaning forward and pointing at her laptop screen, slight smile, soft natural daylight, photoreal UGC iPhone footage aesthetic, screen visible at slight angle showing an inbox dashboard.
Caption overlay: "Until I plugged my inbox into [tool category]. Replies tripled in a week."
Shot 3 (15-20s) — Payoff
Same 32-year-old woman, leaning back smiling, holding her phone toward camera with a sign-up screen visible, soft natural daylight, photoreal UGC iPhone footage aesthetic.
Caption overlay: "Try it free for 7 days. Link in bio."
Three shots, 15 seconds of generated video, identical subject anchor, identical camera language, identical lighting and setting. Total Seedance variable cost on Pro 1080p: roughly $1.86. Stitch in any editor with the captions above plus a light background track. Export 1080x1920 vertical.
Where Brand Brain Fits
This tutorial covers the photoreal-actor format: a synthetic person holding the product, talking to camera, walking through a setting. Brand Brain ships a different but adjacent UGC format natively: AI Avatar talking-head clips. A single still photo (generated in any image model or uploaded) plus voice cloning produces lip-synced vertical clips with locked identity by construction. AI Avatars are included on every paid plan from Starter at $29 per month with 10,000 AI credits.
The two formats compose. Pair Brand Brain AI Avatars for founder explainers and hooks (where identity must be exact across dozens of clips) with Claude-directed Seedance for the photoreal product-in-hand scenes. Brand Brain handles brand voice for captions, multi-platform publishing to LinkedIn, X, Threads, and Instagram, and the content review queue across both formats. The result is a unified UGC pipeline instead of two parallel workflows.
Director. Brief, shot structure, per-shot prompts, captions, narrative review.
Generator. 5-second photoreal clips at 1080p, text-to-video and image-to-video. Native audio on 2.0.
Brand voice, AI Avatar talking-head UGC, content review queue, native publishing to LinkedIn, X, Threads, Instagram.
The trial is 7 days with $1 in starter credits and a card required at signup. That is enough to wire one Claude session, one Seedance ad set, and the AI Avatar layer end-to-end and ship a real spot.
Common Pitfalls
Vague prompts that produce stock-AI faces
Prompts like "young woman talking to camera in a kitchen" return the same generic face every other AI video returns. Anchor the subject with age, clothing, personality cue, and a specific real-world setting. The first time the prompt feels too specific is usually the first time the output feels real.
Mixing UGC and cinematic camera in the same ad
A handheld selfie hook plus a tripod-locked cinematic demo collapses the UGC effect. The viewer subconsciously registers it as a produced ad and skips. Pick one camera language and hold it across all shots in the spot.
No reference frame for character continuity
Three text-to-video Seedance shots of "the same woman" produce three different women. Either generate one image first and use Seedance image-to-video for every shot, or use Higgsfield Soul ID for identity locking. Text-to-video alone does not preserve identity across runs.
Skipping captions
Most paid social plays mute by default. A UGC ad without burnt-in captions for the hook and CTA loses the audience in the first 2 seconds. Caption every spoken line plus the close. Do not rely on platform auto-captions for paid creative.
No attribution tag on the asset
Without a UTM string in the CTA URL or a hidden experiment ID in the filename, the 30-day audit cannot rank which AI-generated variants converted. The workflow stays a hobby instead of compounding into a creative system.
Hitting publish without policy disclosure
Meta, TikTok, and YouTube each require AI-generated content to be labelled when it could mislead viewers about real people, events, or places. Read each platform synthetic media policy and label the asset on upload. The cost of labelling is zero; the cost of an ad ban or account strike is high.
When to Use This vs a UGC Platform
This Claude + Seedance workflow is the right pick when the priority is creative control and per-ad cost, or when the operator already lives in Claude Code. The trade-off is that the operator owns prompt-writing, generation, and stitching as three steps. A managed UGC platform like MakeUGC or Creatify wraps those steps inside one UI with avatar libraries, batch mode, and ad-launcher integrations, at a higher per-month subscription cost.
If the volume is <20 ads per month and the brand voice is non-standard, this manual workflow wins on cost and creative control. If the volume is >50 ads per month and the brand sits in a category the platforms have tuned for (DTC consumer goods, supplements, apparel), a managed platform wins on time. Most teams run a hybrid: managed platform for high-volume variants, Claude + Seedance for hero spots and brand-anchored creative.
Pick Claude + Seedance when: creative control matters, brand voice is non-standard, operator is in Claude already, volume is <20 ads/mo, per-ad cost is the main concern.
Pick a managed UGC platform when: volume is >50 ads/mo, brand sits in a category platforms have tuned for, time matters more than creative control, operator wants ad-launcher integrations.