Make it learn your style before it makes the art: skills, a check pass, then carousels, posts and DMs
Full Value DanX (Twitter) · 2026-09-12
Read originalI'm really surprised with what you can do with @Muse Most agents struggle with making decent images. For marketers, Muse automate can: - Create detailed carousels with custom images - Create skills and verification steps to learn your style - Post to social media - DM people https://t.co/SzTQ84sW1p
Business model
The author says most agents cannot produce decent images, and lists four things the workflow handles: detailed carousels with custom images, skills and verification steps that learn his style, posting to social, and sending DMs. It is a list of capabilities, not a result, with no follower count or conversion anywhere in it, so read it as four stations on one content line rather than a growth story. The one that earns the others is the second: turning style from a feeling into a checklist is what makes batching possible at all. What transfers: feed it five to ten pieces you actually stand behind, make it write the rules as instructions rather than adjectives, check every batch against that list before anything goes out, and send the DMs yourself, one at a time.
Style is the thing everyone claims to have and nobody can hand to a contractor, which is why agent-produced content so often looks competent and anonymous at the same time. Turning it into a checklist is the whole trick, and it is a small one. The other thing worth separating is reach: posting is reversible, sending a private message to a real person is not, so the two belong on different sides of the human gate.
Workflow
- Give it five to ten pieces you genuinely stand behind, not the ones that performed best.
- Make it write what it sees as instructions you could hand to a stranger, never as adjectives.
- Save those rules as a skill with a checklist, and run every batch through it before anything ships.
- Produce one sample slide first, and only expand to the full carousel once it looks like you.
- You decide the timing and the platform, and have it log what went out and how it did.
- Handle DMs separately: read every one and send it yourself.
Adaptation plan
The post gives no follower count, no conversion, no revenue and no example of the content itself. Whether the carousels performed, whether the DMs were answered and how long the setup took is unknown. Image quality and how faithfully a learned style transfers vary a lot by model.
Prompt
AgentFlow starter draft: not tested in the original business setting; results are not guaranteed. Synthetic examples do not verify the author’s revenue.
You are my content and style assistant. Your job is to learn my style first, then produce carousels that pass a check. You never publish anything and you never send a direct message. [INPUT: what I will provide] 1) Five to ten pieces of my old content I stand behind, as style samples: captions, images, headlines. 2) A list of counter-examples, the things that sound unlike me. 3) This round's topic, the target reader and the number of slides, five by default. 4) Brand hard constraints: primary colour, whitespace ratio, banned words and the maximum information per slide. [STEPS] 1. Read every sample and counter-example and write the patterns you see as executable rules: sentence length, vocabulary, primary colour, whitespace, information density. Never leave it at adjectives. 2. Turn those rules into a checklist where every line answers yes or no. 3. Draft the structure and the point of each slide for this topic, and wait for my confirmation. 4. Produce one sample slide. Only after I confirm it sounds like me do you expand to the full set. If it does not, fix the rules first. 5. Run every slide through the checklist, mark failures with a reason and redo them until the whole set passes. 6. If direct messages are needed, list the people and an opening for each, and hand them to me. [OUTPUT] - Style rules, numbered and executable. - The checklist, every line checkable. - The carousel structure: slide number, its point, and the image note. - The check result, per slide, with reasons for anything that failed. - DM candidates, with account and opening, awaiting my confirmation. [CONFIRMATION REQUIRED BEFORE CONTINUING] - The sample slide is confirmed by me before the set is expanded. - Posts and images are published by me; DMs are confirmed one by one and sent by me. - Any figure, case or quote carries its source and date. If you cannot find one, write NOT_OBTAINED. You never invent one. - You never invent follower, conversion or revenue numbers, and you never promise a result. - Mentions of other brands or people are attributed, not rewritten as statements of fact. - The output respects the commercial and copyright terms of the platform it will land on.
Running costs and pitfalls
The post gives no follower count, no conversion, no revenue and no example of the content itself. Whether the carousels performed, whether the DMs were answered and how long the setup took is unknown. Image quality and how faithfully a learned style transfers vary a lot by model. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.
