A creator ad for your product
Turn a product and a two-line brief into a vertical, creator-style talking ad.
Show how to chain a reference mix into a talking avatar. Muse Edit makes one frame with the creator holding the product, an LLM turns the brief into a short spoken script, and LongCat Avatar animates that frame to the voice, so the product stays in her hand for the whole ad.
Saved result
Muse paints a stand-in product: a terracotta self-watering herb pot with basil, sitting in a sage-green reservoir with a round water-level window. It also paints an invented creator in a mustard sweater filming a selfie in a sunny kitchen. Muse Edit puts that exact pot in her hand at chest height in a 9:16 frame, keeping her face, the kitchen and the pot's reservoir and window. GLM writes a 37-word first-person script from a two-line product brief: "I used to kill every basil I brought home. This little terracotta pot sits in a water reservoir, so it drinks on its own, and a tiny window shows me when to refill—like every two weeks. Honestly, get one." MiniMax Speech 2.8 HD reads it in 13 seconds, and LongCat Avatar 1.5 makes her say it at 480p while she holds the pot steady beside her shoulder. Flaws: the prompt asked for shoulder-length hair and she has longer curls. Her expressions run a little big, with a few wide-mouthed frames. The ad is 480p. The script ran 13.1 s against a 10 to 12 s target, so LongCat billed 14 s ($0.56). The editor's estimate for this node follows the voice length (script words before a run, the measured clip after). The creator is an invented person and the ad is AI-generated, so label it that way if you post it. Swap the product Image for a 📷 Image input to use your own product photo. Every still and sampled video frame was checked and is SFW (a clothed invented adult and a plant pot, no text or logos). Inspired by @teddarling's AI Creator Lab for ComfyUI (https://x.com/teddarling/status/2116187911850701017). No media from that post is used.
Another run: A woodworker’s lunchbox adSee what the same workflow made with different inputs ↓
All 280 frames and the spoken script checked against the brief · 2026-10-30 · Reported saved run: $0.52914.
Partial. Same graph, a second input: a sage-green bamboo-fibre lunchbox, a fictional woodworker in his sixties (grey hair and beard, navy apron over flannel) filming a selfie in his workshop, and a calm, warm male voice (MiniMax Patient_Man). Besides the text inputs, the “Hold the product” instruction was adapted to a man and a lunchbox, and the voice was changed; both are listed below. The script is 28 words and the ad runs 11.2 s against a ~10 s brief. It says the box keeps food warm for about four hours and that the silicone seal stops leaks, but it leaves out two brief points, dishwasher-safe and bamboo fibre, so the brief is only partly met. The mouth moves in time with the voice, the lunchbox stays steady by his shoulder in every frame, and there are no cuts, extra people or text. Cost is the NanoGPT balance change across the run: $0.033 for the three Muse images and GLM, about $0.0145 for the speech and $0.4814 for LongCat (11.3 s billed as 12 s).
Okay, this lunchbox is a game changer. It keeps my food warm for about four hours, and the silicone seal means nothing leaks in my bag. Grab one.
Inputs for this version
Product · stand-in
a matte sage-green insulated bamboo-fibre lunchbox with a natural bamboo lid, a silicone seal and a fold-down carry handle, on a light wooden workbench, as a soft daylight product photo.
Creator · look
a fictional cheerful man in his sixties with short grey hair and a trimmed grey beard, wearing a plain navy work apron over a flannel shirt, filming a selfie in a bright woodworking workshop. Invented person, no logos or text.
Hold the product · instruction
Make one vertical selfie-video frame from the two images. Image 1 is the creator: keep this exact man, same face, short grey hair, trimmed grey beard, navy work apron over the flannel shirt, workshop and daylight, same chest-up framing, eyes to camera, mouth relaxed and fully visible. Image 2 is the product: he holds exactly one of this lunchbox in one hand at chest height beside his shoulder, showing it to the camera, with the same matte sage-green body, natural bamboo lid, silicone seal and fold-down carry handle. The lunchbox must not cover his face, beard or mouth. One person, one product, no other copies, no text, no logos, no stickers.
Product brief
the lunchbox keeps food warm for about 4 hours, is leak-proof thanks to the silicone seal, can go in the dishwasher, and is made from bamboo fibre. Aim for about 10 seconds and 28 words.
Delivery
Relaxed and friendly selfie-video delivery, with small nods. Natural blinks and a calm, warm smile between sentences. He keeps holding the lunchbox steady beside his shoulder; the lunchbox and his hand stay the same. Face and mouth stay visible, eyes toward the camera, handheld phone framing with only slight natural sway. No cuts, no extra people, no text.
Voice · voice
Patient_Man
Sampled models: meta/muse-image/text-to-image, meta/muse-image/edit, z-ai/glm-5.3-flash, Minimax-Speech-2.8-HD, longcat-avatar-1.5.
Make it yours
- Open the graph — save first if the canvas already has your work. Undo brings yours back.
- Rewrite Product · stand-in for your product, or swap the Product shot Image for a 📷 Image input with your own product photo. Put what it really does in Product brief, because the script only claims what the brief says. Change Creator · look for a different invented creator or setting, keep "face and mouth fully visible" in Hold the product · instruction so the lip sync has a clear mouth, and set the tone in Delivery. Pick another voice on the Voice node.
- Check the editor estimate, then run with your NanoGPT key. Muse 1:1 $0.011 + Muse 9:16 $0.011 + Muse Edit 9:16 $0.011 + GLM 5.3 Flash under $0.001 + MiniMax Speech about $0.02 + LongCat Avatar 1.5 at $0.04 per second of voice at 480p, billed in whole seconds ($0.56 for 14 s). The saved run was $0.614. The editor prices LongCat off the voice length: the script's words (about 2.8 per second) before a run, the measured clip after one. When the script length is unknown (nothing wired into Speech yet), the estimate assumes 40 s of voice. LongCat bills at least 3 s and at most 30 s of voice, so the chip clamps the seconds first and then prices them: $0.12–$1.20 at 480p ($0.04/s) and $0.24–$2.40 at 720p ($0.08/s). A second saved input (a woodworker's lunchbox, 11.3 s of voice billed as 12 s) cost $0.529.
- Check that the selfie frame keeps your product's shape and details and doesn't cover her mouth, read the script for claims the brief doesn't make, and watch the ad with sound to judge the lip sync and whether the product stays steady.
Download workflow · Original sampled graph · Explore MCP tools · awesome-noodles