AI image workflows that hold up

AI image workflows that hold up

Image models are extraordinary cameras with unreliable taste. The creators who look “lucky” are usually running a repeatable stills workflow. This page is that workflow for product, editorial, concept, and social stills—plugged into the site-wide creative workflow system.

What good looks like

A professional still usually has:

• Clear subject hierarchy (you know what to look at in half a second)
• Believable light direction
• Intentional crop for the final placement (hero, square, story)
• Few or zero artifact taxes that take longer to fix than regenerate
• Finish: grade, type, and cleanup outside the generator when needed

Pretty is not the bar. Usable in context is the bar.

Workflow A — Hero still (marketing / web)

1. Brief the frame

Answer:

• Aspect (16:9 hero vs 1:1 ad)
• Subject + action
• Environment
• Light (window soft / hard noon / studio softbox)
• Brand adjectives (calm, loud, clinical, handmade)
• Forbidden list (neon, stock handshake, cluttered logos)

2. Build a prompt in layers

Subject: matte black pour-over kettle on oak counter
Setting: modern kitchen, morning
Light: soft window light from camera left, gentle shadows
Lens mood: 50mm, shallow depth of field
Style: photoreal product editorial
Palette: warm oak, cool stainless, muted greens
Composition: kettle left third, negative space right for headline

Add avoid list: busy background, harsh flash, readable brand text, extra handles, warped metal.

3. Generate a small batch

Produce 3–4 candidates. Pick by silhouette readability at mobile width, not only by texture porn at full zoom.

4. Critique against placement

Mock the still into a rough webpage or ad frame. Many “winners” die when type lands on a busy corner.

5. Refine

• Inpaint steam, reflections, or edge glitches
• Outpaint if crop is tight for responsive layouts
• Keep seed/composition when only light or material should change

6. Finish

Export → light grade → remove sensor-dust-like spots → add real typography in a design tool. Do not ask the model to spell your brand unless you enjoy proofreading disasters.

Workflow B — Style-consistent series (campaign set)

Clients rarely need one image. They need a set that feels related.

1. Lock a style paragraph (light + medium + palette) reused every prompt.
2. Lock camera height and lens mood.
3. Vary only subject or scene.
4. Prefer tools with reference / character / style controls when available.
5. Build a contact sheet of six; kill outliers that break the set.
6. Finish as a batch with the same grade preset.

If frame three looks like a different movie, the set is not a campaign—it is a collage.

Workflow C — Product accuracy (real object)

When the object must match reality (packaging shape, device ports, furniture proportions):

1. Start from real photos of the product.
2. Use img2img / product reference features at moderate strength.
3. Prompt for scene and light, not inventing the object.
4. Never trust the model to invent accurate UI screens—composite real UI or keep screens off.
5. Legal: do not fake endorsements or alter safety-critical product details.

AI fantasy kettles are fine for mood. AI “exact SKU” claims are not.

Workflow D — Concept art and exploration

Exploration allows messier Stage 3:

1. Wide prompt families (three style directions).
2. Larger batch for direction only.
3. Art director picks one direction.
4. Then switch to production rules (small batches, one variable).

Timebox exploration (30–60 minutes). Endless concepting is procrastination with receipts.

Composition controls worth learning

When available, pose, depth, or edge control beats rewriting the same prompt twenty times hoping geometry improves.

Iteration pattern that actually teaches you

If you change light and lens and style every time, you learn nothing.

Worked example — quiet tea brand hero

v1 prompt (compressed): ceramic cup on linen, morning light, photoreal.

Result: Nice cup, but café neon sign in background—brand fail.

Critique: Kill neon; keep cup angle and linen texture.

v2: same subject, “minimal home kitchen, soft north window light, no signage, muted sage palette,” avoid neon, logo, text.

Finish: crop for 16:9, slight warmth, headline set in Figma on negative space.

Archive the “no neon” rule in the brand negative list forever.

Artifacts checklist (stills)

• Hands and fingers (if any body parts appear)
• Text and labels
• Reflections that invent extra objects
• Edges that melt into background
• Over-smooth skin plastic if people appear
• Inconsistent shadows

If fix time exceeds ten minutes, regenerate with tighter constraints.

Finish stack recommendations

Generators do not know your ad account’s crop rules. You do.

Tool choice (short)

Use the AI creative tools scorecard for weighted picking. Rough defaults:

• Aesthetic concepting → strong aesthetic models
• Client commercial comfort → tools with clearer licensing stories + Photoshop finish
• Privacy / custom style → local or fine-tune friendly stacks

Ethics and rights (short, important)

• Do not deliberately imitate a living artist’s name as a style shortcut for commercial work without understanding the risk.
• Do not generate trademarked characters or celebrity likenesses for brand campaigns.
• Disclose AI use when a client, platform, or contest requires it.
• Rights details vary by vendor—read terms; this site is not legal advice.

More in later rights pages; until then, when money is on the line, ask counsel.

Next

1. Run Workflow A on a real asset you need this week.
2. Save prompts + seeds.
3. Continue to AI video workflows if motion is next.
4. Compare tools with the scorecard.

Published by Tabaconda LLC, Florida, USA.

Related tools & gear

Useful physical tools for image production work. Optional product searches (see disclosure). Buy only what you need.

• Color calibrator
• Gray card
• Tripod