Short AI ads: creators report $7–$9 production costs

• Short AI-ad cost claims clustered between $7 and $9: iamneubert reported $9 in 10 minutes in iamneubert’s $9 ad, while c_valenzuelab reported $7 in nine minutes in c_valenzuelab’s $7 ad.
• The documented $9 setup used Runway Agent in quality mode, and iamneubert said in iamneubert’s quality-mode reply that Agent selected Seedance 2.5 for the video.
• A repeatable ad structure can generate variants across eight comic settings, according to iamneubert’s concept screen, and a knight version was another $9, per iamneubert’s follow-up.
• A separate previs workflow put one Seedance 2.5 generation at about $10, while HalimAlrasihi said in HalimAlrasihi’s cost breakdown that its 3D previs took three to five minutes.
• Agent editing can complete basic cuts, but moritzkremb called an early DaVinci Resolve result basic in moritzkremb’s DaVinci test.

Runway’s Agent documentation says the chat-based tool can start from a product, image, campaign, or idea, then plan and produce a project. Its developer model catalog lists seedance2_5 among the available video models.

The three posts give quick output, time, and cost figures, rather than a common production budget.

• iamneubert: $9, 10 minutes iamneubert’s $9 ad
• c_valenzuelab: $7, nine minutes c_valenzuelab’s $7 ad
• c_valenzuelab: $8, 9.5 minutes c_valenzuelab’s $8 ad

Neither creator specifies source-asset preparation, rerolls, total clips generated, or a definition of what the figure covers in those posts. iamneubert later said in iamneubert’s production reply that the $9 result was a single generation with a logo added.

iamneubert described a short Agent brief in iamneubert’s workflow reply: provide the idea, desired model, and relevant campaign details, then let it develop creative directions. For the $9 video, the creator said in iamneubert’s quality-mode reply that Agent ran in quality mode and generated with Seedance 2.5.

Runway says users can choose whether Agent waits for approval and what it optimizes for when selecting models in its Agent guide. iamneubert also said in iamneubert’s model-selection reply that Agent chose the model for that run; Runway’s separate Model Router documentation describes an API route configured around cost, latency, or quality.

iamneubert shared a chat in iamneubert’s concept screen that keeps one comic structure intact: relaxed close-up, an urgent “one more quick change” ping, an alarmed look around, then repeated zoom-outs showing how far the character is from a laptop.

The proposed settings were:

1. Ski lift
2. Hot-air balloon
3. Ferris-wheel top car
4. Zipline
5. Desert camel trek
6. Rock-climbing wall or cliff face
7. Wedding altar
8. Sauna or ice bath

The same reply says Claude or ChatGPT can generate more ads that preserve the structure and flow, and that Runway’s MCP or plugin can handle the process inside chat. iamneubert then reported another $9 generation featuring a knight in iamneubert’s knight variant.

runwayml showed a more layered route in runwayml’s product-animation workflow: make a style frame in Runway from a product image, animate in Blender, then render in Seedance 2.5. HalimAlrasihi used a comparable division, GPT Astra plus Blender for blocking, camera movement, framing, and timing, then Seedance 2.5 plus references for the finished scene in HalimAlrasihi’s previs workflow.

HalimAlrasihi put the Blender previs at three to five minutes and one Seedance generation at around $10 in HalimAlrasihi’s cost breakdown. The creator said in HalimAlrasihi’s restyle explanation that a whole previs can be uploaded to Seedance 2.5 for transformation or restyling, adding light, elements, movement, or different characters.

In a separate Astra test, moritzkremb gave the agent DaVinci Resolve access. It retrieved files, started a project, and made cuts and overlays, but the creator called the initial output basic in moritzkremb’s DaVinci test.

The tester said in moritzkremb’s editing follow-up that cuts worked reasonably well, while overlays and animations were harder. A later Hyperframes run identified a sequence from raw footage, made cuts, and added text and overlays, according to moritzkremb’s Hyperframes result.