The one-person creator stack: Claude, Midjourney, Kling, ElevenLabs

Use this stack as a page: the one-person creator stack.

A year ago “AI content” meant one chatbot that wrote a caption and one image model that made a square. In 2026 the people actually publishing — product launches, YouTube essays, a weekly Short — run a pipeline. Four tools, four jobs, one artifact moving down the line.

The stack that stuck for a one-person studio:

1. Claude (or ChatGPT if that is already your writing home) — the script.
2. Midjourney — the stills that set the look.
3. Kling AI — the motion.
4. ElevenLabs — the voice you can ship.

This is not a “top 50 creator tools” list. It is how a 45-second product video gets out the door without a crew. Image roundups and voice roundups already exist on this site; this post is the handoffs.

Why one generator for everything broke

Text models can describe a shot. They cannot hold a character’s face across twelve frames. Image models can lock a look. They cannot speak. Video models now emit native audio (Kling VIDEO 3.0 does dialogue, sound, ambience) — and that audio is still the wrong tool for a brand voice you will reuse for six months.

If you ask one product to write, illustrate, animate, and narrate, you get a clip that looks like a demo. The stack exists so each step has a specialist, and so you can re-run one step without regenerating the other three.

ChatGPT is fine for the script if that is where you already think. We default to Claude here because the writing job is long-form control and a brief that will be reused as image prompts and shot lists — the same reason we pick Claude for essays in Claude Opus 5 for writing. Swap ChatGPT if your voice is already there. Do not run both on the same script.

The four roles (and what each must not do)

Claude owns the brief and the script

Claude is the reasoning / long-context assistant. Free row plus paid from $20/month on our matrix. You are not using it as an image model. You are using it to produce:

• A 45-second script with timestamps.
• A shot list (what is on screen for each line).
• Image prompts that match the shot list (same character, same lighting language).
• A voice direction note for ElevenLabs (pace, not a novel).

Our matrix marks Claude’s context window as 1M+ tokens. That is the reason to keep the product page, the brand notes, and the last three scripts in one thread instead of starting from zero every Monday.

Do not ask Claude to “just make the video.” There is no video file at the end of that sentence. Do not ask it to clone a voice. That is ElevenLabs, and cloning someone else without consent is still a scam vector — we said the same in ElevenLabs, ChatGPT Voice, and Suno.

Midjourney owns the look

Midjourney is paid (from $10/month). There is no free consumer tier on our matrix. Default model in 2026 is V8.2. Use the website; /imagine in Discord is leftover muscle memory. Details: How to use Midjourney V8.

Your job in Midjourney is not “a cool picture.” It is a consistent still you can feed to Kling as a start frame:

• --ar 9:16 for Shorts / Reels, --ar 16:9 for YouTube.
• --sref or Personalization (--p) so shot 2 matches shot 1.
• The Edit Model when you need to change one object without rolling a new character.

Do not generate a new hero face for every shot. Kling will treat each still as a different person. Do not use Midjourney as your video tool. That is Kling (or Runway, if you swap).

If you need in-image text (a pack shot with a readable label), Ideogram is the specialist we track. Use it for that still, then come back. It is not a fifth member of the default stack.

Kling owns the clip

Kling AI is Kuaishou’s video and image studio. VIDEO 3.0 / VIDEO 3.0 Omni: text-to-video, image-to-video, start/end frames, native audio, multi-shot control up to 15 seconds. The web app has a free credit pool; paid memberships start at $10/month on our matrix. API billing is a separate prepaid Open Platform — not the consumer plan.

This is the motion step. You take the Midjourney still as a start frame (image-to-video is a documented VIDEO 3.0 path) and a prompt that is the shot list line, not a poem. Clip length is 3–15 seconds. A 45-second video is several clips, not one generation. That is the whole craft: stitch, do not pray for a feature film.

Native audio on the clip is useful for ambience and throwaway dialogue. It is not your brand narrator. Our matrix marks Kling voice mode as no — native audio is generated on the clip, there is no live conversational loop. Leave the voiceover to ElevenLabs.

Do not skip Midjourney and text-to-video everything unless you have accepted that the character will drift. Do not upload a PDF of the script; Kling’s file upload row is no. The prompt is the file.

ElevenLabs owns the voice you reuse

ElevenLabs is commercial TTS: emotive speech, cloning, dubbing. Free row plus paid from $5/month. This is the studio narrator, the product demo voice, the thing you will still want in October.

Handoff: the Claude script (clean, no shot directions in the spoken lines) → ElevenLabs → a WAV you lay under the Kling cuts. If you need music, that is Suno or Udio, and those are not this stack. Mixing a generated song under a product demo is a different video.

Do not use ChatGPT voice mode to dub a 45-second ad. Voice mode is a live conversation loop (plus-tier on ChatGPT). Wrong job. Do not clone a celebrity “for the vibe.”

Walkthrough: one 45-second product video

Named job: a vertical ad for a water bottle — problem, product, three features, call to action. No talking head. No crew.

1. Claude: script and prompts

Paste the product page and three constraints: 45 seconds, 9:16, no claims you cannot back. Ask for:

1. Spoken script (what ElevenLabs will say), with timings.
2. Shot list aligned to those timings (6–8 shots; remember Kling tops out at 15s per clip).
3. One Midjourney prompt block that shares character, lighting, and lens language, then varies only the action.

Handoff artifact: a single markdown note. If the spoken script still contains “[show the cap]” in the voiceover line, edit it. ElevenLabs will read the stage direction out loud.

2. Midjourney: the stills

Run the prompt block. Lock the hero bottle and the lighting on shot 1. Use --ar 9:16. For shots 2–6, vary the action, not the product. If the bottle mutates, stop and use the Edit Model rather than rolling a new seed.

Handoff artifact: 6–8 stills in a folder, named 01-hook.png, 02-pour.png, … not mj-dump-final-FINAL.png. Kling’s start-frame step needs you to know which image is which shot.

3. Kling: motion in pieces

For each still: image-to-video, 5–8 seconds, camera move from the shot list (slow push, not a helicopter). Generate two takes. Pick one. Repeat. Stitch in whatever editor you already have — CapCut, Descript, Premiere. The stack does not include an NLE on purpose. Adding Descript is a swap if your whole workflow is transcript-based; it is not required for this ad.

If a shot needs a start and end pose (bottle closed → cap off), use start/end-frame. That is a documented Kling path. Do not describe the cap in prose and hope.

Handoff artifact: a silent (or ambience-only) picture-lock of the 45 seconds. Picture-lock before voiceover. If you voice first, you will recut to the voice and then regenerate video to match, forever.

4. ElevenLabs: the take

Paste the spoken script. Generate two takes. Pick the one that fits the picture-lock without you speeding the video up to 1.25×. Lay it under the cut. If a line is long, cut the line in Claude and re-render the voice — do not stretch Kling clips past 15s with a prayer.

Handoff artifact: the export. You are done when the CTA is readable and the bottle still looks like the same bottle.

What actually moves between tools

• Product page → Claude: constraints plus claims you can stand behind.
• Claude → Midjourney: a prompt block with shared look language.
• Midjourney → Kling: named stills as start frames, one shot each.
• Claude → ElevenLabs: the spoken script only — no shot directions.
• Kling + ElevenLabs → editor: picture-lock plus a WAV.

If you skip the named stills and dump a Midjourney zip into Kling, you will lose the shot list. The stack fails at organization more often than at models.

Cost, from our matrix

Solo-creator paid floor if you take every paid row:

• Claude: free to start, $20/month when you need the paid assistant.
• Midjourney: $10/month (no free consumer row).
• Kling: free credits, then $10/month.
• ElevenLabs: free to start, $5/month when you outgrow it.

You can cheap out on Claude (stay on the free row for a 45-second script) more safely than you can cheap out on Midjourney. There is no free Midjourney tier on our matrix; that is the look. Kling’s free pool is for tests, not a weekly ad calendar.

For a free-image detour, top free image generators and the Flux / Ideogram follow-up exist. A Flux still can replace Midjourney in this pipeline. The handoff to Kling does not care which model made the PNG. It cares that the PNG is consistent.

When to swap a piece

Claude → ChatGPT if that is already where you write. Same job. See /compare/chatgpt-vs-claude. Do not dual-write the script in both.

Midjourney → Flux / Ideogram if you need a free or typography-heavy still. The rest of the stack stays.

Kling → Runway if you want Runway’s editor / VFX-style control more than Kling’s start-frame and native-audio clip factory. Runway has a free row and paid from $12/month. Side by side: /compare/runway-vs-kling-ai. Pick one video generator per project. Generating the same shot in both and “seeing which you like” doubles cost and breaks continuity.

Kling → Veo 3 if you already pay for Google AI Pro and want native audio inside Google Flow / Gemini. Veo is the Google-native swap, not a second video tab.

ElevenLabs → ChatGPT voice only for a live working session, never for the shipped dub. ElevenLabs → Murf / Descript if the team already edits in that suite. The stack’s point is a reusable studio voice, not a new narrator every week.

What not to add

• Sora plus Kling plus Runway on one ad. Three video models is not quality control. It is a slot machine.
• Canva Magic Studio as a fifth “just to resize.” If you need a thumbnail, export a still from Midjourney. Canva is a real product (/tools/canva-magic-studio); it is also how this pipeline turns into a junk drawer.
• A music generator in the default path. Add Suno when the video is a song. A product ad needs a voice and room tone more than a chorus.
• HeyGen / Synthesia talking heads on top of Kling b-roll unless the format is a spokesperson. Those are a different stack (avatar, not motion-from-stills).

What to actually pick

• Default solo studio, 2026: Claude + Midjourney + Kling + ElevenLabs.
• Already in Google: Claude or Gemini for the script, Midjourney or Flux for stills, Veo for motion, ElevenLabs still for the brand voice (Veo’s native audio is not a voice library).
• Budget stills: Flux or Ideogram instead of Midjourney; do not drop ElevenLabs to save $5 if the voice is the product.

Capability pages: /tools/claude · /tools/midjourney · /tools/kling-ai · /tools/elevenlabs.