# Midjourney V8.2 Lab: First Controlled Frame Midjourney can create an attractive image quickly. Understanding why it looks that way and which part you can repeat is harder. Add an image, moodboard, style reference, and a handful of parameters at once, and your conclusion is usually “something worked.” That is not enough for learning. In this lab, you will create **one baseline frame**, change **one variable**, and compare real results. The goal is not the greatest number of settings; it is a clear reason for the next decision. ```learner-callout {"stable_id":"c190-key-baseline","role":"key-point","title":"Start with a baseline frame","body":"Lock V8.2, obtain one checkable result without a style reference or moodboard, then change only one style mechanism or parameter. Two comparable files teach more than one overloaded request."} ``` By the end of the lab, you will have a version record, compact prompt, two images or the honest status “prepared to run,” and a decision about which variant to accept and why. ## 1. What Midjourney Is and Where the Prompt Goes Midjourney is an image-generation service. You can use the Midjourney web interface available to you or a Discord route. A prompt is an image description submitted in an image-creation field. The model version determines which parameters and reference routes are compatible with the task. | Item | Role in the work | | --- | --- | | website or Discord | where the task is submitted | | Midjourney | the image-generation service | | V8.2 | model version for this lab | | prompt | description of the required frame | | parameter | short command at the end of a prompt, such as ratio or version | Open the current official interface, find the image-creation field, check the selected version, and submit one prompt. Button names and setting locations can change. ```learner-callout {"stable_id":"c190-source-freshness","role":"source","title":"V8.2 is a dated route","body":"The local adapter is statically validated for V8.2, but runtime is not proven. Technical facts require a new official check when Version or compatibility documentation changes or freshness expires; current availability and result are checked separately."} ``` ## 2. First Run: from Empty Field to Frame Begin without references or a moodboard. The practice task is a vertical product frame: one dark green glass bottle on the right, with empty space for a heading on the left. 1. Open the current Midjourney site or your working Discord route. 2. Confirm that V8.2 is selected, or add `--v 8.2` at the end of the prompt. 3. Paste the test prompt and submit it. 4. Open the variants at a useful size. 5. Accept only a frame with one bottle, an empty left third, and readable glass edges. ```prompt luxury product photograph, one dark green glass bottle on a pale stone surface, the bottle placed in the right third, clean negative space across the left third, hard side light from the left revealing the thick glass edges, muted forest green and warm gray palette, restrained editorial finish --ar 4:5 --raw --s 100 --v 8.2 ``` `--ar 4:5` sets a vertical ratio, `--raw` reduces automatic styling, `--s 100` is a baseline stylization strength, and `--v 8.2` locks the version. `4:5` and `100` are choices for this test, not universally best values. ```learner-callout {"stable_id":"c190-observation-baseline-check","role":"observation","title":"A beautiful second bottle still fails","body":"Check the task itself: one object, bottle in the right third, real empty area on the left, and light that reveals the glass thickness. If one criterion fails, reject the baseline test."} ``` ## 3. Why the Version Must Be Recorded Parameters and workflows change between versions. Keep this record beside every test: ```template Version: V8.2 Interface: website / Discord Check date: Official Version page: Settings actually available: ``` On August 11, 2026, official Midjourney Version material described V8.2 as the default and gave July 24, 2026 as its release date. This is a historical source for the lesson, not a replacement for today’s check. Before a run, consult the current Version page: an old parameter can look convincing while belonging to another version. ## 4. One Main Method for Controlling Style The baseline result first shows whether the text task works. Only then add **one** source of visual language: a text style dictionary, one Style Reference, or one Moodboard/personalization profile. An Image Prompt may be added separately when content or composition influence is needed, but it does not prove exact identity. ```learner-callout {"stable_id":"c190-technique-one-style-engine","role":"technique","title":"One style engine in a baseline test","body":"Do not stack a Style Reference, Moodboard, high Stylize, Weird, and Exp at once. Choose one primary mechanism first, record its code or source, and keep the other meaningful conditions unchanged."} ``` ### Style Reference A Style Reference transfers broad visual character: palette, material, texture, light, and overall look. It is not a way to copy a person, object, or scene content. ```template [final-image description] --sref --sw 100 --v 8.2 ``` Text description is mandatory. For V8.2, do not add `--sv`; `--sw` starts at `100` as a test baseline, not a guarantee. ### Moodboard A Moodboard is a broader project style through `--p `. Record the exact code and test date. ```template [final-image description] --p --v 8.2 ``` For the first comparison, choose Style Reference **or** Moodboard, not both. A Moodboard does not combine with `--sw` or `--sv`. ## 5. What V8.2 Does Not Promise In the currently statically validated V8.2 adapter, do not use `Character Reference`, `Character Weight`, `Omni Reference`, `Omni Reference Weight`, `--q`, `::`, `--sv`, `--oref`, `--ow`, `--cref`, `--cw`, `--niji`, `--turbo`, or `--style raw`. Use `--raw`, not `--style raw`. Image Prompt and Image Weight can assist content, composition, and color influence, but do not guarantee an exact face, product geometry, packaging, or logo. When exact identity is a hard criterion, choose a verified route that can actually preserve it or add the exact element using a controlled tool after generation. Draft Mode has a conflict in historical official material: Version-table and Draft/Conversational documentation disagreed for V8.2. Draft is therefore not part of the mandatory lesson route; before important use, check the current interface and both pages. ## 6. Describe the Frame Describe the final image in ordinary language first: ```text type of image -> number of objects and main subject -> their frame positions -> viewpoint -> lighting direction -> important materials and textures -> required empty area -> verified parameters in use ``` Be concrete: `clean negative space across the left third` is stronger than “some room.” For light, name source, direction, hardness, and contrast when they matter. For `--no`, use separate concrete nouns or attributes, not ambiguous phrases. ## 7. Text, Logos, and Exact Packaging This lesson does not promise exact Cyrillic, a logo, label, or packaging. When lettering is a hard criterion: 1. prepare a composition with a clean area; 2. generate the background or foundation; 3. add exact text with a controlled tool; 4. do not regenerate an accepted design because of one wrong letter. This is a fallback to an exact result, not a claim that the service never creates text. Short Latin text may be more reliable, but exact typography still needs verification. ```learner-callout {"stable_id":"c190-caution-exactness","role":"caution","title":"Resemblance is not exactness","body":"Do not describe a similar face, product, package, logo, or text as proven exact. When the criterion is hard, protect a clean area, check the file, and add the exact element through a controlled route."} ``` ## 8. Second Test: One Variable Return to the baseline bottle and change one layer—for example, add one Style Reference. Keep object, composition, light, ratio, Raw, Stylize, and version the same. ```template Baseline file: Second-test file: Conditions kept the same: One changed variable: What changed in color, texture, or light: What broke accidentally: Accepted result and reason: ``` ```learner-callout {"stable_id":"c190-inference-one-variable","role":"inference","title":"A comparison answers one question","body":"If you change reference, ratio, light, Raw, and the entire text together, two attractive images will not explain the cause. A recorded version, a seed only as a repeatability aid, and one delta create a usable comparison."} ``` If the interface or compatibility changed, stop the run, record the actual version, open the official Version page and selected-feature page, then remove unconfirmed syntax. ## 9. Practice and Lab Completion 1. Choose your own image task or the bottle example. 2. Record interface, V8.2, and check date. 3. Produce a baseline result without a style reference or moodboard. 4. Save the accepted file or record why variants were rejected. 5. Add one style engine or change one parameter. 6. Produce the second result and compare it with the baseline. ```learner-callout {"stable_id":"c190-practice-honest-status","role":"practice","title":"Preparation is not generation","body":"When images were produced, compare them with the card and accept one variant. If access or execution is unavailable, keep the honest status “version and two prompts prepared; generation not run.”"} ``` ### Check before Leaving ```learner-callout {"stable_id":"c190-checkpoint-v82-ready","role":"checkpoint","title":"The lab is completed in a checkable way","body":"There is an actual version and date, a baseline and one variant with one changed variable, two files or an honest no-run status, a visible acceptance criterion, and one next correction. A V8.2 prompt ends with `--v 8.2` and contains no unconfirmed legacy syntax."} ``` Takeaway: a controlled frame begins not with the maximum number of parameters but with one clear baseline result. Produce it, change one variable, and let two real images show what worked.