Flow is Google's filmmaking environment built around Veo-class models. Its practical value is organization, not just generation: reference/ingredient-led creation, scenes assembled from beats, and extensions that continue a shot instead of re-rolling it. Exact features, model names, and credit costs change — recheck at lab time.

Prerequisites

• 59.1 brief reduced to two beats (hook + proof, or proof + close).
• Approved ingredients: 1–3 reference frames (character, product, style/location) plus any text direction.
• A 59.A card; note plan, resolution, and per-clip settings before generating anything large.

Workshop path

1. Create the project; add ingredients. Upload approved references with names. Record which inputs the current Flow surface accepts. 2. Generate the first 4–8 second scene. One beat, one camera move, motion-anatomy direction. Save version, settings, cost. 3. Extend or build the next beat with the current extension/continuation feature rather than a fresh random roll. The goal is deliberate continuation: same world, next action. 4. Arrange clips in a scene. Order beats, trim heads/tails, check visual and auditory continuity (light, wardrobe, product, direction, ambient tone). 5. Export per the brief. Vertical or wide as planned; recheck safe areas after any reframe. 6. Log the cost/control decision. Model choice, resolution, clip length, and feature use all affect credits — record what each second actually cost and whether a cheaper setting would have served the beat.

Cost awareness (part of the craft)

• Check current credit/pricing display before batch generation.
• Generate short and few; extend winners rather than re-rolling losers.
• Prefer locked cameras and short durations for proof shots — they cost less and edit better.
• Record date, plan, and per-clip cost in the 59.A card so next month's pricing does not surprise you.

Worked mini-example + troubleshooting

Two beats, one world: beat one, a 5-second wide of the bakery at dawn (ingredients: location still + style frame); beat two, an extension into the baker scoring dough at the counter — same light, same apron, camera pushing in slowly. If the extension breaks continuity (apron color shifts, window jumps sides), shorten the extension, re-anchor the character reference, and state screen direction explicitly ("baker frame-left, window light from left, movement left → right"). If costs spike, drop resolution on exploratory versions and reserve full quality for the selected beats only. Save ingredients, both directions, versions, continuity notes, and the per-second cost line — that record is the lab's real output alongside the scene itself.

Scene-assembly checklist (before export)

• Beats in story order with trims: no model wobble at heads/tails, no repeated gestures across the join.
• Continuity holds: light direction, wardrobe, product, screen direction, ambient tone consistent beat to beat.
• Audio intent set: narration or bed placed, levels roughly balanced, captions planned (full caption pass happens in 59.H / Lesson 59.5 style).
• Ratios decided: vertical or wide per the brief, safe areas rechecked after any reframe — never blind auto-crop.
• Versions labeled: flow-beat01_v2-WINNER.mp4 style, so the 59.A card points at files, not memories.

When Flow/Veo is the wrong tool

Say so in the log when true: a single locked product close-up may be cheaper and faster in an image-to-video specialist; a dialogue-heavy explainer may belong in an avatar/editor environment; a shot needing pixel-exact UI should be a real screen capture, not a generation. Flow earns its place on multi-beat, reference-led scenes with continuations — route other jobs elsewhere without guilt. That routing judgment, written with costs attached, is the lab's passing grade.

Verify fast

Play beat one into beat two muted. One continuous world? Product and wardrobe intact? Then confirm the cost line: can you state the per-usable-second price and the cheaper fallback? If not, the lab is half done.