A creator documents a $9 Seedance 2.5 ad workflow with Runway Agent

• The $9 and 10-minute headline describes one reported 30-second output, which iamneubert's original post presents as fully generated.
• The reusable asset is a timed production brief, not a vague style prompt: iamneubert's full prompt locks character continuity, sound, camera motion, and an end frame for a tagline.
• Runway Agent handled the first creative pass, then the creator turned its narrative structure into a prompt that can produce new situations, as iamneubert's workflow breakdown explains.
• A separate Seedance workflow from HalimAlrasihi's Blender previs workflow uses GPT Astra and Blender to lock blocking, framing, and timing before the video model renders the finished scene.

The prompt specifies a notification ping at 0.8 seconds, a voice note from 1.2 to 4 seconds, then four continuously widening camera moves. Runway's Seedance 2.5 guide confirms that a native 30-second run is a supported format, rather than a stitched sequence of short clips.

The creator reports $9 and 10 minutes for the finished ad. They later said the video was “100% generated” in iamneubert's reply, and described this particular output as a single generation with a logo added in iamneubert's single-generation reply.

The result is a reported outcome, not a promised rate. They added “Not guaranteed” in iamneubert's caveat.

The published prompt uses six blocks:

1. Character: a tightly specified woman, outfit, yellow Jeep, location, and an instruction to keep those details identical for all 30 seconds.
2. Style: live-action 16:9, 35mm grain, late-afternoon Utah light, deadpan timing, and a single continuous take.
3. Audio: ambient desert sound, a phone ping at 0.8 seconds, then a short “one more quick change” voice note.
4. Timeline: a close-up reaction followed by four escalating pull-backs, ending with the Jeep as a tiny point in the landscape.
5. Camera rules: level horizon, the Jeep near the center, eased movement, and no jump cuts between scales.
6. Negative constraints: no additional people or vehicles, roads, buildings, overlays, subtitles, night, or weather shifts.

The final 3 seconds hold on an extreme wide frame, leaving empty visual territory for a campaign line.

The initial brief gave Runway Agent a simple comic skeleton: someone relaxing on vacation receives a boss's urgent final-change text, looks worried, and is revealed through a huge zoom-out to be nowhere near a laptop. It also asked for a super-wide end frame for taglines.

After a result lands, the creator asks the agent for situations with the same structure. iamneubert's scenario list returned eight swaps:

1. A stopped ski lift.
2. A hot-air balloon over a distant launch field.
3. A Ferris-wheel car at the apex.
4. A zipline over a canyon.
5. A camel trek in the dunes.
6. A climber on a sheer face.
7. A groomsman interrupted at a wedding altar.
8. A remote sauna or ice bath.

The creator says they usually use Runway's web app in iamneubert's reply, but their breakdown also names the ChatGPT Runway plugin and Runway MCP as routes for the same agent-led workflow. Runway's MCP setup guide says the connector lets ChatGPT, Claude, and other compatible LLM agents generate images and video inside a conversation.

For this run, iamneubert says they selected quality mode and Runway Agent generated the video with Seedance 2.5 in iamneubert's model reply. A second post asked the agent for a knight version of the format.

Runway's MCP page publishes the connector address, https://mcp.runwayml.com/mcp, for compatible-agent setups.

Halim Alrasihi documented a different pipeline for scenes where camera geometry and timing are set before generation:

1. GPT Astra makes a Blender blockout for camera movement, framing, and timing.
2. The creator asks GPT Astra for a scene breakdown, including action and timing, then uses those details in the video prompts, according to HalimAlrasihi's reply.
3. The complete previs goes into Seedance 2.5 as a video reference, which can transform or restyle it with lighting, elements, motion, or new characters, as HalimAlrasihi's restyle explanation describes.

Alrasihi reported 3 to 5 minutes for the GPT Astra previs and roughly $10 for one Seedance 2.5 generation in HalimAlrasihi's cost reply.

Runway calls Seedance 2.5 a ByteDance third-party model and lists it as available in Tools, Workflows, and Agent in its model documentation. It accepts text, image, video, and audio inputs, supports 4 to 30 seconds, and can use up to 50 references.

Its listed base generation rates make the 30-second format easy to price in credits:

• 480p: 20 credits per second, 600 credits for 30 seconds.
• 720p: 30 credits per second, 900 credits for 30 seconds.
• 1080p: 68 credits per second, 2,040 credits for 30 seconds.

A settings screenshot posted by iamneubert shows the 30-second, 720p configuration at 900 credits per video.