A tiered model workflow routes AI image generation tasks across model tiers by purpose: a cheap, fast model — Nano Banana 2 Lite at 3 cents per image — handles high-volume exploration and drafts, and only selected or technically demanding assets escalate to a premium model like Nano Banana Pro for final production quality.
A tiered model workflow splits your image pipeline into two roles: an exploration tier where you generate in volume without budgeting individual images, and a production tier reserved for the assets that actually ship. The routing rule is simple — everything starts cheap, and only what you select moves up.
The exploration tier. Run all mood boards, concept tests, ad drafts, and pre-visualization frames on a low-cost model. Nano Banana 2 Lite generates an image in about 4 seconds for 3 cents — 1,000 images for $30, one-quarter the cost of Nano Banana Pro and 2.5x faster than Nano Banana 2. Its 1K resolution is a deliberate design trade-off for this role, not a quality failure: at this price, generating 10 options instead of one becomes the rational default, and Google's own benchmarks show the budget tier now outperforming the premium tier on text-to-image quality — the cheapest model is no longer the worst model.
The escalation tier. From the exploration batch, shortlist the winners and re-run only those on Nano Banana Pro — final production output and technically hard shots that demand the highest quality. Nothing else touches the expensive tier. That single rule is what produces the cost saving: at 1,000-image volume, Lite runs four times cheaper than Pro, so the fewer images that escalate, the more the workflow pays off.
The routing layer. The practical failure point in tiered workflows isn't model quality — it's losing creative context across dozens of generations. invideo is an agentic creation platform with the current image models (Nano Banana 2 Lite, Nano Banana Pro, GPT-Image-2, Recraft) available in one place, and the invideo agent solves the context problem with persistent memory: load your character sheets, brand context, and creative direction once, and every subsequent generation in a high-volume session pulls from that context automatically while the invideo agent routes each task to the appropriate tier — no rebriefing, no picking a platform per model.
Where the pattern pays off. D2C advertisers use it to test 50 ad concept variants instead of 5 — for less than the cost of a coffee — then escalate the winning creative, directly reducing customer acquisition cost. Filmmakers use it for pre-production: iterating on 20 mood boards used to be a budget conversation and is now practically free on the exploration tier. Agencies run all mockups and draft rounds on Lite, effectively eliminating internal draft review costs, and escalate only client-approved selects to Pro.
Watch some of these to see what works for you:
only the ones that are chosen get escalated to Nano Banana Pro
— invideo's creative team