AI painting and design nanny-level tutorial: from prompt words to commercial delivery

🛒 For the design, operation, e-commerce and self-media people, a complete drawing guide from registration tool to delivery document package.

Tutorial objectives and applicable readers

This tutorial teaches you step by step how to use AI drawing tools to produce usable design drawings and marketing drawings from scratch: registering the tool, writing prompts, creating concept drawings, refining, typesetting, and commercial inspection. No art or programming background is required, designers, operators, e-commerce, and self-media creators can get through just by following these steps. The entire journey takes approximately 60-90 minutes to complete.

1. Preparation Checklist

  • [ ] Sign up for Midjourney (high-quality concept art, subscription-based) or choose the cloud-based Stable Diffusion service
  • [ ] Register Canva (free file is enough, used for typesetting and multi-size adaptation)
  • [ ] Register Ideogram (free file, suitable for images with text)
  • [ ] Prepare 2-3 brand reference pictures (color, layout, style) as picture input
  • [ ] Clarify the first picture requirement (it is recommended to choose a poster or social media picture)
  • [ ] Prepare a quick check of brand guidelines (main color values, fonts, logo usage rules)

2. Tool registration and entry

  1. Open the Midjourney official website to register and subscribe to the basic file, and enter the Discord server (or use the official web version).
  2. Open Canva to register a free account, and become familiar with the "AI Image Generation" and "Batch Creation" entrances.
  3. Open Ideogram registration and become familiar with how to generate posters with text.
  4. Create an "AI Drawing" folder with four subdirectories: 01-Concept, 02-Refinement, 03-Typesetting, and 04-Delivery, which correspond to the assembly line links one by one.

3. Five elements of prompt words

Translate "I want a good-looking picture" into a language that AI can understand, and fill in the blanks with the following five elements:

Subject: {Describe the core object/character/scene}
Style: {illustration/realistic/3D/flat/national trend/cyber...}
Light: {natural light/neon/soft light/backlight...}
Composition: {center/rule of thirds/close-up/wide angle...}
Parameters: {ratio --ar 16:9, degree of stylization --stylize, etc.}

Example (Midjourney):

Subject: An orange cat programmer wearing headphones sits in front of the screen with coffee on the table
Style: flat illustration, warm colors
Light: Soft indoor light
Composition: Center, half-length close-up
Parameters: --ar 16:9 --stylize 200

Writing prompt words well and being specific is the key first step to improve the quality of drawings by 80%.

4. Create a concept map: divergence and screening

  1. Enter the prompt word into Midjourney to generate 4 variations at a time.
  2. Use the Upscale and Variations buttons to continue deriving in the desired direction.
  3. For a picture that is not satisfied with the whole but likes the part, use local redraw (inpaint) to change a certain area individually, for example:
Replace the main body with an office building and keep the rest unchanged
  1. Save 2-3 pictures in each round into 01-Concept, and mark the direction with the file name (such as concept-flat-cat-01).
  2. The goal of the concept stage is to be "many and complete". Don't get hung up on the details first and let AI help you exhaust all the style possibilities.

5. Controllable refinement: drawings and reference drawings

After selecting the direction, switch from "random" to "controllable":

  1. Import the selected concept drawing as a reference/underlay drawing.
  2. Use image-to-image to adjust the style, light, and color while retaining the main structure.
  3. When it is necessary to strictly lock the composition, use the ControlNet class capability (Canny/Posture) to constrain the picture structure; if the cloud service does not have this function, use multiple drawings to gradually approximate the picture.
  4. Use the redraw tool to repair parts that are not satisfactory (hands, text, edges) individually.
  5. Enlarge the base image you are satisfied with, export it to a high-definition version, and save it in 02-Refinement.

6. Typesetting and multi-size adaptation

  1. Open Canva and import the refined base image as the background.
  2. Add brand text and logo, apply brand color matching, and complete layout.
  3. Use Canva’s “Resize/Batch Create” function to export multi-platform sizes (official account cover, Xiaohongshu 3:4, Moments 1:1, Douyin 9:16) with one click.
  4. If you need a picture with complex text, you can import it into Canva after the Ideogram is generated and then typeset it.
  5. The finished product is saved in 03-Typesetting, and the file is exported to mark the channel and size.

7. Commercial inspection and delivery

  1. Authorization Check: Confirm that the model used is allowed for commercial use (Midjourney paid version/Canva/Firefly, etc. can be used for commercial use; free files and specific models need to check the terms one by one).
  2. Watermark Check: Confirm that there is no platform watermark in the output image and no AI-generated residual mark (if it needs to be removed, it needs to be regenerated with an authorized version).
  3. Brand Check: Check whether the color matching, logo, and fonts are used correctly against the brand specifications.
  4. Screen Quality Inspection: Check hand, text typos, and edge abnormalities one by one, and return problematic pictures for refinement.
  5. Archive 04-Delivery, attach the "Authorization Statement + Source File + Export of Each Size" file package.

8. Verification method

  • Map output speed: from prompt word to usable base map ≤10 minutes
  • One-time pass rate: ≥3 out of 5 consecutive pictures can go directly to typesetting without major changes.
  • Brand consistency: the main color/font/style of the same series of images is consistent
  • Commercial compliance: Spot checks on delivery pictures show no watermarks and clear authorization
  • Complete delivery: each set of materials includes source files and full-size export

9. Frequently Asked Questions and Troubleshooting (FAQ)

  1. Hands and fingers are always drawn wrong? Use partial redrawing to repair the hands alone, or change to a model/version that is better at human body; the prompts describe posture constraints such as "hands naturally placed on the table".
  2. Chinese text is always garbled? It is recommended to use Ideogram or add words in Canva to generate Chinese text, which is more reliable than letting the image model draw Chinese directly.
  3. The generated image was judged not to be commercially available by the platform? For commercial projects, priority is given to products marked as commercially safe, such as Firefly, Canva, and Midjourney paid versions, and check the authorization terms.
  4. The style of the same set of pictures is not consistent? Fixed model version, prompt word template and reference image, use style LoRA or unified seed when necessary.
  5. The prompt word is always a little off? Use the five elements of "subject + style + light + composition + parameters" to self-examine and make up for what is missing; read more excellent prompts of the same style to accumulate corpus.
  6. The computer cannot bring the local model? Start with cloud subscription or online services to avoid the threshold of local deployment, and then evaluate self-build when the volume is large.

10. Advanced expansion

  • ComfyUI workflow: Solidify common drawing processes into node templates, team sharing, batch drawing
  • Brand fine-tuning: Use a small amount of brand materials to fine-tune LoRA so that the AI stable output fits the brand's style.
  • API Content Factory: Access the image generation API to realize the automation of "copywriting → automatic image matching → multi-size export"
  • Linked Writing: Cooperate with the AI writing pipeline to achieve a closed-loop content production integrating graphics and text

11. Practical example: Complete rendering of a social media poster

Take "New Product Launch Promotional Poster (1:1 Square)" as an example to link up the entire process:

  1. Disassembly brief: Main body = new smart desk lamp; style = minimalist Nordic style, warm light; composition = desk lamp in the center, white space; text layer = product name + one sentence selling point; size = 1:1.
  2. Concept: Enter "smart desk lamp, minimalist Nordic style, warm light, centered composition, half body, --ar 1:1" in Midjourney, generate 4 variations, and then derive the variations of your favorite image points.
  3. Refinement: Select one piece, use a drawing to warm the light, and remove excess elements. If you are not satisfied with the details of the desk lamp base, use partial redrawing to repair it separately.
  4. Enlarge and export: Enlarge the satisfactory base image to 2048px and export a high-definition version.
  5. Typesetting: Import Canva, add product name and selling point text, brand color embellishment, and apply logo specifications.
  6. Multiple sizes: Use "Adjust Size" to batch export official account covers, Xiaohongshu 3:4, and Moments 1:1.
  7. Commercial Check: Confirm that the Midjourney paid version can be used commercially, has no watermark, has complete font authorization, and is archived and delivered.

Remember the parameters and prompts of this set of steps as your "poster template". You only need to change the subject and copy for the next picture, and you can have a usable first draft within 5 minutes.

User Reviews

  • Loading reviews...