4.5Combined workflow (Midjourney, Photoshop, Firefly)
The essentials in 30 seconds
- No single tool: a four-step chain — create (Midjourney), fix (inpainting, Photoshop, Firefly), compose (logo, type), finalize (color, upscale, export).
- Don't ask the AI for what another tool does better: text, alignment, exact brand color and logo are added after, exact and editable.
- Sonia's reflex: work non-destructively (originals, layers, prompts, versions) and document the chain that works — to reproduce, vary and delegate.
Running case — Sonia and the café campaign, end to end
Sonia, freelance graphic designer in Algiers, must deliver the café's full campaign: storefront poster, web banner and social posts — same mood, same logo, same typeface, three formats. She creates the base with Midjourney, planning the right format and free space for text from generation time. She fixes flaws (inpainting for localized ones, Photoshop and Firefly for seams), composes in her graphic tool (logo, typography and brand elements, exact and editable), then finalizes (shared color harmonization, upscale per resolution, exports). She keeps originals, layers, prompts and versions — and writes her chain sheet. Last milestone of the running case: Sonia no longer produces images, she runs a chain.
Last case of the 4.1-4.5 running case — Sonia completes her expert level.
1. A four-step chain
Professionals don't look for one unique tool: they build a chain where each tool does what it does best. The most common workflow follows four steps. 1) Create with the strongest aesthetic tool — typically Midjourney — to quickly get a visually strong base: composition, light, mood. 2) Fix and complete: inpainting for localized flaws (section 4.1), Photoshop and its Generative Fill or Firefly for retouches and additions (section 4.4), frame extension if needed (Pan / Zoom Out).
3) Compose: assemble with the other elements — logo, typography, brand elements, other images — in Photoshop, Illustrator, Figma or Canva depending on the medium. 4) Finalize: color harmonization, upscale to the required resolution, export to the medium's formats. Like Sonia's café campaign, each link has its role and nobody does another's job.
2. Don't ask the AI for what another tool does better
The principle that saves the most wasted time: typographic text, precise alignment, an exact brand color, a logo — all this is added after, in a tool where it is editable and exact (section 3.2). Instead, generate the image with a free space planned for it.
Start from the deliverable, not the tools: what to produce (banners, illustrations, print)? At what frequency and volume? With which constraints (brand, formats, privacy, budget)? The answers set the chain. Then start minimal — one generation tool, one retouch/compose tool — and only add links (upscaler, ControlNet, LoRA) for real need.
3. Non-destructive and documented: industrializing
Work non-destructively and keep intermediate steps: keep the original generated image, your layers, prompts and versions. That lets you vary, fix late, and reproduce the series later — like the three formats of Sonia's campaign from one shared base.
And document your chain: once you find the sequence that gives good results (such prompt type → such retouch → such upscale → such export), write it down. You will save major time, get consistency across productions, and be able to delegate or industrialize — exactly the next level's topic. The most important tool stays your eye: selection and art direction matter more than any software.
🔎 Concrete example — 🔎 Concrete example — The café campaign, three formats, one chain
One base, three deliverables. Sonia's sequence:
1. Midjourney base: right --ar per medium + free space for text.
2. Fixing: inpainting (localized) → Photoshop/Firefly (seams, additions).
3. Composition: exact, editable logo + type.
4. Finalizing: shared color grading, per-resolution upscale, exports — layers and prompts archived.Reading: create, fix, compose, finalize — each tool in its place.
Lesson: a documented chain turns tinkering into reproducible production.
Section vocabulary
- Combined workflow
- Tool chain where each acts on what it does best: create, fix, compose, finalize.
Simple example: Sonia's café campaign: Midjourney, then Photoshop, then composition, then exports.
What it is not: doing everything — including text and logo — in the generator alone.
Reflex: new project: draw the chain (4 links) before opening any tool.
- Creation phase
- Generating the visual base (composition, light, mood) with the aesthetically strongest tool.
Simple example: Sonia generates the café's mood base with Midjourney, text space planned.
What it is not: fixing flaws or adding the logo.
Reflex: creation: aim for mood and composition, not detail perfection.
- Non-destructive work
- Keeping originals, layers and versions to fix and vary later.
Simple example: Sonia produces poster, banner and posts from the same archived base.
What it is not: overwriting the original at each step with no version kept.
Reflex: each step: keep the previous version before moving on.
- Chain documentation
- Writing down the sequence that works, to save time, stay consistent and delegate.
Simple example: Sonia's chain sheet: prompt type, retouches, upscale settings, export formats.
What it is not: a series of different tinkerings for each visual.
Reflex: working chain: write it down at once, with prompts and settings.
In practice — In practice — Setting up your production chain
- Create a strong base with Midjourney, planning the right format (--ar) and free space for text from generation time.
- Fix flaws: inpainting for localized ones, then Photoshop/Firefly for seams and additions.
- Compose in your graphic tool: add logo, typography and brand elements (exact and editable).
- Finalize: color harmonization, upscale to the medium's resolution, export; keep originals, layers and prompts.
Ready-to-say script — presenting your chain to a client (30 s)
“This is Sonia: for your campaign, I create the mood with AI, I hand-fix what must be fixed, I add your logo and text pixel-exact, and I produce the three formats with the same color grading. Each step is archived — we can vary or fix anytime without starting over.”
Say this when defending a quote: detail the 4 links to justify the finishing price.
Copy-paste template — production chain sheet
Project: [café campaign — poster + banner + posts]
1. CREATE: Midjourney — prompt type [formula] — --ar per medium — text space [position]
2. FIX: inpainting [areas] → Photoshop/Firefly [areas]
3. COMPOSE: logo [file], type [name], colors [references] — [tool]
4. FINALIZE: color [shared setting] — upscale [tool/setting] — exports [formats]
Archives: originals + layers + prompts + versions [folder]Multi-format campaign
Like Sonia: one base, three variants, shared color grading — the chain pays off from the second format.
Single demanding visual
Same chain, concentrated: explore, fix, compose, finalize — without skipping verification.
Regular production
Prompt type + chain sheet + archives: each new visual reuses the machinery — time halved.
Work to delegate
The chain sheet + archives let a second designer take over without starting over.
✓ Best practices
- Create the base with the aesthetically strongest tool, finish elsewhere.
- Add text, logo and brand colors after generation — exact and editable.
- Work non-destructively and document the chain as soon as it works.
✗ Common pitfalls
- Asking the AI for typographic text or an exact brand color.
- Cropping for lack of planning format and free space from generation time.
- Not keeping originals, layers and prompts — impossible to vary or fix after.
Key takeaways
- TYPE CHAIN: create (Midjourney) → fix (inpainting, Photoshop/Firefly) → compose (logo, type) → finalize (color, upscale, export).
- DON'T ask the AI for what another tool does better: text, alignment, brand colors are added AFTER.
- PLAN free space and format from generation time to avoid cropping and losses.
- NON-DESTRUCTIVE (originals, layers, prompts, versions) + DOCUMENTED CHAIN to reproduce and delegate.
48h challenge — apply tomorrow
On a real visual to deliver: apply the 4 links (create with text space, fix, compose logo + type, finalize + export) and write your one-page chain sheet.
- I can chain create → fix → compose → finalize without skipping a step.
- I add text, logo and brand after generation, exact and editable.
- I archive originals, layers and prompts and document my chain.
Frequently asked questions
Which tool should I start with to build a production chain?
Start from the deliverable, not the tools: what to produce, at what frequency and volume, with which constraints (brand, formats, privacy, budget)? Then: creation (Midjourney if aesthetics rule, local Stable Diffusion if privacy, volume or control are critical), fixing (Photoshop as reference, Firefly for simple), composition (Photoshop, Illustrator, Figma or Canva per your practice). Start minimal, optimize the bottleneck, document from the start. The most important tool stays your eye.
How long does it really take to produce a professional visual with AI?
Far less than before, but more than demos: generation takes seconds, but exploration (often longest), fixing, composing, finalizing and checking follow. A well-mastered simple visual: tens of minutes; a demanding visual: hours. The gain stays massive on exploration and variants. Two factors: your mastery (light and framing vocabulary) and your preparation (documented chain, brand, prompt types). If you bill: sell art direction, selection and finishing.
What is the key principle of a combined workflow?