One clean product photo is enough to build a UGC-style ad on an 8frame board: Nano Banana stills put the product in a creator's hand and room, an image-to-video node animates each beat from its still, and the Video Editor adds captions and sound. The chain is upload, scene stills, one video node per beat, timeline. The limit people do not expect: the Montage node that joins clips on the canvas drops all sound, so a talking or voiced ad has to be assembled on the Video Editor timeline instead.
For the beat structure of the script, see the UGC ad template; for planning shots first, how to storyboard a UGC ad. This page stays on the canvas, node by node.
TL;DR
| Step | Node or model on 8frame | Fixed price (credits) | Input it needs |
|---|---|---|---|
| 1. Product photo | Upload | none | your best front photo |
| 2. Clean cut-out (optional) | Background removal | 12 per image | the product photo |
| 3. Scene stills, one per beat | Nano Banana 2 | 11 per image at 1K, 17 at 2K | product photo as reference, prompt |
| 3b. Hero still with tighter label | Nano Banana Pro | 21 per image at 1K or 2K, 42 at 4K | product photo as reference |
| 4. Motion per beat | Seedance 2.0 (720p) | 108 per 5 s clip, 48 at 480p | one start still |
| 4b. Motion, longer beat | Seedance 2.5 (720p) | 157 per 5 s clip | one start still |
| 4c. Motion, cheaper draft | Kling v3 Standard | 17.01 per second with sound, 11.34 silent (86 / 57 for 5 s) | one start still (required) |
| 5. Voiceover (optional) | ElevenLabs TTS | 13 per started 1,000 characters | the script text |
| 6. Captions and sound | Video Editor | not a generation | the clips, the script |
The board, node by node
There is no published template for this exact chain yet, so you build it on a blank board. The closest live starting point is PRODUCT PLACEMENT, which handles step 3 (putting a product into a scene). You can open it, keep its scene step, and add the video and voice nodes below.
Step 1: the product photo
Upload the best photo you have: front-facing, even light, label readable, nothing cropped. The canvas upload takes common image files up to 50 MB but not iPhone HEIC, so export to JPG first. The photo is the reference every later node looks at: a window reflected in a glossy bottle will follow it into every scene.
Step 2: cut-out (only when the background fights you)
If the product sits on a busy table, run Background removal (12 credits per image). A clean cut-out gives the scene model less to misread. Skip it when the photo is already on a plain backdrop: the scene model can read the product from the original.
Step 3: one still per beat
A UGC ad is a handful of beats: hook, problem or context, product in hand, result, call to action. Make one still for each beat, each in its own Nano Banana node, all fed from the same product photo.
8frame has no batch node and no variable injection, so each still is its own node, and each can be re-run alone when one beat is off.
Nano Banana 2 is the default Nano Banana model, 11 credits per image at 1K, 17 at 2K, and it takes up to 10 reference images. Use it for the creator-in-a-room beats. Switch one node to Nano Banana Pro (21 at 1K or 2K, 42 at 4K) for the product hero beat, where the label has to be read at full size.
A prompt for the "product in hand" beat:
Phone-camera selfie angle, a woman in her thirties in a bright kitchen holds the bottle from the reference photo at chest height, label facing camera, natural window light from the left, slightly imperfect framing like a real creator video, vertical 9:16
Keep the person fictional and generic. No real or famous faces, and if you use your own face or a team member's, get written consent for ad use first.
What to check on each still: the label text against the real photo, cap and colour, the size of the product in the hand compared with a real hand, and the light direction, which should match across all beats so the cut does not jump.
Step 4: one video node per beat
Connect each approved still to its own video node as the start image. Here is how the three main choices behave on the canvas:
- Seedance 2.0 at 720p is 108 credits for 5 seconds (48 at 480p), runs 4 to 15 seconds, and includes sound. Its price does not change when you add reference images.
- Seedance 2.5 at 720p is 157 credits for 5 seconds and runs 4 to 30 seconds, so a long product demo beat can be one take. A 30 second take is 937 at 720p, not 157.
- Kling v3 Standard takes any whole second from 3 to 15, billed at 17.01 credits per second with sound or 11.34 silent (86 or 57 for 5 seconds, 171 for 10 with sound). It is image-to-video only on 8frame and follows the start still's shape, which is fine here because every beat has a still.
Seedance tops out at 720p and Kling v3 Standard returns 720p. For vertical social ads that is the working resolution; there is no video upscaler on 8frame, so a beat that must be 1080p goes on Kling 2.6 Pro (95 for 5 seconds with sound, 47 silent, measured 1080p output) from the start.
Prompt the motion, not the scene. The still already holds the scene.
Handheld phone camera, she tilts the bottle toward the lens and smiles, small natural camera shake, no cut, the label stays facing camera
Step 5: voice (optional)
If the ad has a voiceover rather than on-camera speech, write the script and run ElevenLabs TTS. It bills 13 credits per started 1,000 characters, up to 5,000 characters per generation. A 300 character script bills 13, and a 1,200 character script bills 26, because the second block has started. Eleven v3 and Multilingual v2 cost the same.
For an on-camera creator who speaks, see the lip-sync nodes compared in AI lip sync price by model. That is a different chain (lip sync from a still portrait), not image-to-video.
Step 6: assemble on the Video Editor timeline
Open the Video Editor mode, add the clips in order, drop the voiceover on a sound layer, and use "Generate captions": it turns a pasted script into a caption layer timed at an estimated 160 words per minute. It does not transcribe audio and does not export SRT, so paste the exact words you used in TTS and nudge timing by eye.
Do not use the Montage node for this step. Montage joins 2 to 10 clips with transitions for 10 credits, but the merged file has no audio track: the clips' own sound and any dialogue are dropped. Montage suits a silent loop, not a UGC ad.
Which model for which step
| Need | Model | Length on 8frame | Resolution | Ratios and refs | End frame on canvas |
|---|---|---|---|---|---|
| Creator scene stills | Nano Banana 2 | still | up to 4K | up to 10 references | n/a |
| Label-critical hero still | Nano Banana Pro | still | 1K, 2K, 4K | up to 10 references | n/a |
| Beat motion, short | Seedance 2.0 | 4 to 15 s | 480p, 720p | start image; refs do not change the price | yes, with a start image and no reference images |
| Beat motion, long take | Seedance 2.5 | 4 to 30 s | 480p, 720p | start image | yes, same rule |
| Cheaper draft motion | Kling v3 Standard | 3 to 15 s | 720p | start image required; follows its shape | no |
| Sharper hero beat | Kling 2.6 Pro | 5 or 10 s | 1080p | start image; own aspect setting | no |
The 3 second floor on Kling v3 Standard matters for UGC: a hook or a reaction beat rarely needs 5 seconds, and you pay per second.
Where it breaks
- Sound disappears in Montage. The merged file has no audio track. Assemble on the Video Editor timeline.
- Label drift in motion. A label that is correct on the still can warp when the hand turns the product. Prompt for the label to stay facing camera, keep motion small, and re-run the beat rather than the whole ad. Why AI changes your product label goes deeper.
- Close-up faces in Seedance. Seedance 2.0 and 2.5 refuse reference images with a large photoreal close-up face (error E005); waist-up and full-body references pass; retrying does not help. Frame the creator waist-up with the product visible, which is the UGC look anyway.
- One error message for most failures. Almost every failed run shows "Sensitive content detected. Please try changing your prompt or image." It does not always mean the content was sensitive. Failed runs are refunded automatically; change the still or the prompt and try again.
- 720p ceiling. Seedance and Kling v3 Standard deliver 720p and there is no video upscaler. If a placement demands 1080p, that beat goes on Kling 2.6 Pro.
Platform rule: the AI label
TikTok's ad policy (Misleading and false content, last updated April 2026) allows AI-generated and significantly edited media when you apply its AIGC label or add your own clear disclaimer, and rejects or restricts AI content that is not disclosed. A fully AI creator holding your product is exactly that case: switch the label on in Ads Manager.
Meta labels ads made or significantly edited with its own generative tools, and its post updated June 1, 2026 says Meta will also begin detecting ads made with third-party AI tools. Plan for an "AI info" label.
The same TikTok policy also bans misleading before-and-after product comparisons, so a generated "after" shot is the riskiest beat on this board. Keep results beats to what the product visibly is.
Why this route is in demand
According to a Quiver Quantitative summary of a Fiverr release of June 9, 2026, searches on Fiverr for "AI UGC video ads" rose 265% between May to October 2025 and November 2025 to April 2026. That is search interest on a freelance marketplace, not ad performance.
FAQ
Can I make a UGC ad from one product photo without filming anything? Yes. The product photo is the only real input. Scene stills, motion and voice are generated per node, and the creator in the frame is fictional. Disclose it as AI where the platform asks.
Which video model should I start with? Seedance 2.0 at 720p (108 credits for 5 seconds) for most beats, Kling v3 Standard (17.01 per second with sound, 11.34 silent) for cheaper and shorter drafts, and Kling 2.6 Pro (95 for 5 seconds) only for a beat that must be 1080p.
Why did my ad lose its sound? You probably joined the clips with the Montage node. Montage outputs no audio track. Rebuild the sequence on the Video Editor timeline and add the voiceover on a sound layer.
Do I have to label the ad as AI? On TikTok, yes, for content like this: the policy says undisclosed AI content will be rejected or restricted. Meta adds its own "AI info" label in many cases. Check each platform's page before launch.
Sources
- 8frame prices, defaults and limits: the 8frame app, checked 2026-11-09.
- TikTok Ads Policy, Misleading and false content (last updated April 2026): https://ads.tiktok.com/help/article/tiktok-ads-policy-misleading-and-false-content, checked 2026-11-09.
- Meta, "Expanding GenAI Transparency for Meta's Ads Products" (published 2025-02-03, updated 2026-06-01): https://about.fb.com/news/2025/02/gen-ai-transparency-metas-ads-products/, checked 2026-11-09.
- Quiver Quantitative summary of the Fiverr Business Trends Index (2026-06-09): https://www.quiverquant.com/news/Surge+in+Demand+for+Claude+Code+Specialists+on+Fiverr+Highlights+Shifting+Business+Focus+Towards+AI+Automation+and+Content+Creation, checked 2026-11-09.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.