Should I generate images before videos to save AI credits?

Yes — generate the still image first, lock the framing, then spend video credits only on the locked frame. Images cost a fraction of video generations, and across documented productions only 9–26% of generated video clips made the final cut, so any framing decision you can settle at the image stage saves real credits.

Treat every shot as image-first: generate a still, iterate on composition, lighting and subject until the frame is right, then animate only that locked frame. The invideo agent is built for this — it routes still generations to image models (GPT-Image-2, Nano Banana, Recraft) and only sends locked frames to video models like Seedance 2.0, Kling or Veo, so you're not paying video rates to discover that the framing was wrong.

The numbers make the case. One documented UGC ad generated 11 video clips and used 1 (9% utilization). Another generated 39 video clips and used 10 (26%). Across a 4-ad run, 108 images and 103 videos were generated and 49 clips made final cut. Image generation also carries a 65% discount on invideo's generative plan, so running 10 image variants to find the right framing costs a small fraction of one video regeneration. As Hridaye, invideo's creative director, puts it: "I only spent video credits on locked frames."

The workflow, step by step:

1. Generate a still of the shot first. Ask the invideo agent for the frame as a static image — character, garment/product, location, pose, lighting all in one image. Iterate cheaply: reframe, swap lighting, adjust pose at image cost, not video cost.

2. Run a single probe shot before any batch. Pick one representative frame — character, garment, set, skin, pose all in it. If that holds, the system is ready to scale. Skipping the probe risks propagating the same failure across every clip you generate.

3. Validate the product at three distances. For product or fashion shots, confirm the item renders consistently as a close, mid and wide image before committing video credits. If the product drifts across image distances, it will drift worse in motion.

4. Lock the frame, then animate. Once the still is approved, pass it to the video model as the keyframe. The invideo agent uses Seedance 2.0 reference-to-video for this — your locked still becomes the anchor, so the video inherits framing, character, and color rather than rerolling them.

5. Batch video in 5s, not 15s. Generate animated clips in batches of 5 rather than the full shot list at once — smaller batches catch drift early and burn fewer credits than discovering a problem 15 clips deep.

6. Lock and regenerate. Keep the clips that land; regenerate only the ones that fail. Never regenerate the whole batch when one shot drifts.

A couple of cases where you don't need image-first: a simple single-location, single-character UGC ad can often jump straight to keyframe iteration without a full character/location sheet — and if you're using Seedance 2.0's single-pass multi-shot generation to preserve audio continuity across cuts, you're committing to one render anyway. Outside those, image-first is the default.

Documented end-to-end costs with this discipline: ~$73 and ~$130 per UGC ad, ~$67 for a 30–35s real estate B-roll package (vs $300–500 for a traditional drone shoot), and ~$150 for a full editorial campaign of 40 stills + 30 motion clips — all including rejected generations counted in the spend.

Watch some of these to see what works for you:

I only spent video credits on locked frames.

— Hridaye, invideo's creative director