Clarinet’s Troubleshooting Midjourney Text Prompts
Discord Contact: @whatnostop (clarinet)
plz send any positive feedback, i need that dopamine!
Sharing this document? Here’s a non-spammy URL for you: https://bit.ly/Clarinet-Prompt-Troubleshooting
Last updated by @whatnostop (clarinet) 6:00 PM CST 10/25/2023
This is not a hard rule, but a soft suggestion that many people use to improve their prompting.
The Early Bits | The Middle Bits | The Last Bits | The End Bit |
Subject | Other Details & Surroundings | Stylizations, Media Type, Artists | Parameters |
a botanical-bearded fairy prince, flowing hair, sky-eyes, symmetrical mossy antlers, intensely sad gaze, wearing a floral diadem, | magical details, twilight atmosphere, | in the style of ArtGerm, Alyssa Monks, Studio Ghibli, close-up, glamour shot | --v 5 |
The Early Bits | The Middle Bits | The Last Bits | The End Bit |
Subject | Other Details & Surroundings | Stylizations, Media Type, Artists | Parameters |
Word order is more nuanced in v4. It doesn’t matter as much. That said,if you need to start somewhere, see the v3 template. | --v 4 |
The Early Bits | The Middle Bits | The Last Bits | The End Bit |
Subject | Other Details & Surroundings | Stylizations, Media Type, Artists | Parameters |
a botanical-bearded fairy prince, flowing hair, sky-eyes, symmetrical mossy antlers, intensely sad gaze, wearing a floral diadem, | magical details, twilight atmosphere, | by ArtGerm, Alyssa Monks, Studio Ghibli, close-up, glamour shot | --v 3 --aspect 9:16 |
The Early Bits | The Middle Bits | The Last Bits | The End Bit |
Stylizations, Media Type, Artists | Subject | Other Details & Surroundings | Parameters |
ArtGerm, Alyssa Monks, Studio Ghibli, close-up, glamour shot, | a botanical-bearded fairy prince, flowing hair, sky-eyes, symmetrical mossy antlers, intensely sad gaze, wearing a floral diadem, | magical details, twilight atmosphere | --test --creative --aspect 2:3 |
Output of v3 example: https://www.midjourney.com/app/jobs/49c34178-5067-4f29-bcc6-5a2671d425a1/
Output of test/testp example: https://www.midjourney.com/app/jobs/07d72d6b-26f0-48df-a568-176f3791e9a4/
(Multi-prompts should belong to only one category and then sorted into that category.)
Replace prepositional phrases with evocative adjectives.
🌐 What Is a Prepositional Phrase? | Grammarly
Replace vague prepositional phrases with evocative verbs.
You may see improved results when you replace vague words with specific words.
This method adjusts Midge’s focus and may increase coherence.
Look for places where your adjectives are already baked in. Eliminating a token reduces the number of combinatorial outcomes, which may increase coherence.
Use words that carry the role/context of the stuff you want to see.
Words with more “baggage” do more work for you.
Use archetypes or stock characters. 🌐Archetype - Wikipedia
Learn more on Discord: https://discord.com/channels/662267976984297473/1024347588792684584
Your prompt may need more chances to be expressed. If you've only used /imagine once and stopped at the first grid, your prompt has had exactly one chance to express itself. “One chance” doesn’t work very well in a system of combinatorics and permutations. You need to give your prompt more than just one chance. Make grid selections to guide it toward the look you're chasing (vrolls), or, if you dislike all the options, 🔄️reroll the whole grid. After 3-5 🔄️rerolls or vrolls, if you see that nothing looks right, it's finally time to troubleshoot the text of your prompt and start again. YMMV. |
When the --quality command is available, you might consider using higher values. When you use higher values of quality, you’re giving Midge permission to “bake” things for a longer time in the GPU oven. It's spending your GPU minutes. The more GPU minutes you spend, the better the output's quality. Not aesthetic quality, but technical quality. If the best your prompt can produce is incoherent garbage, you'll get to see that incoherent garbage at its highest technical quality.😆
Q VALUE | WHAT YOU’RE SAYING WITH IT |
0.25 | I want a very quick sketch of something this prompt might produce. |
1 | I need higher quality because I want a clear picture of what this prompt might do for me. |
2 | I want you to spend some decent time rendering this prompt so that the result is clear, detailed, and aesthetic. |
5 | I want the best quality you can give me from this prompt. |
Learn more on Discord: https://discord.com/channels/662267976984297473/1029192765176823849
HEY! Just use punctuation in Midjourney the way you learned in school. Ok, that said, when troubleshooting or resolving issues, the following information may be useful.
Hi. In this section, I show you how to mess around with scenes and themes, but only because using weights to create themes and scenes is a fun way to learn weights.
Sentience::3 This is a picture of a chaotic mess of external wiring and plumbing on the walls of a brutalist architecture building in a busy city evening atmosphere. The colors are mainly midnight-blue and oak-brown. --ar 5:9
Twilight, vintage coffee house::1.5 Hipster female with blue eyes and long flowing brown hair poses wearing a floral tank and jeans, photorealistic edge-to-edge print in the style of Joe Ranft, Tom Ruegger, John Canemaker, Atelier Lulua --ar 5:7
Rainy forgotten jungle ruins:: Two elderly male explorers argue about credit cards, monsoon atmosphere, photorealistic edge-to-edge print in the style of Joe Ranft, Tom Ruegger, John Canemaker, Atelier Lulua --ar 5:7
If you notice Midjourney appears to "ignore" parts of your prompt, it's time to learn the slider method! This is when you create a promptlet out of an exact phrase (an exact phrase) in your main prompt, and give it a virtual "slider" 🎚️ you can adjust to increase/decrease influence.
1️⃣ First create and run a base prompt like this:
A green dog sleeps in a forest clearing, magical twilight atmosphere --ar 3:2
2️⃣ And then, if you notice a detail is lost, pick it up exactly (exactly) as written from the main prompt and add a slider like this:
A green dog sleeps in a forest clearing, magical twilight atmosphere:: magical::0.3 --ar 3:2
Note that the sliders are put on exact phrases from the main prompt. And the further down the prompt they are, the higher the slider is set to compensate for their later (weaker) position in the prompt.
⚠️ For sliders to work, you need a main prompt that is all ONE string/sentence/statement. It can't have weights already. If your main prompt already has weights in it, this method won't work.
Learn more on Discord: https://ptb.discord.com/channels/662267976984297473/1086302697789083758
Midjourney defines things in terms of their most dominant standard characteristics. Departures from the standard mean Midjourney doesn’t recognize it anymore. One method to invoke an object that breaks its archetype is to figure out what tokens allow for it. For example, you can’t get “a woman with green skin” but you can get “a green-skinned female humanoid” because, to Midjourney, “a woman” can not have green skin per the archetype for humans, but a humanoid can have green skin per the archetype for non-humans.
Another excellent way to break out of archetypes is to use negative weights.You may need to use negative weights to fight stereotypes. For example, if you’re struggling to give something a “non-standard” feature you can try giving a negative weight to the feature you DON’T want to see.
In this example below, we want to give a cat a non-standard third eye. We give Midge permission to add a third eye by replacing the “cat” archetype with “felinoid” and steered Midge away from two eyes by giving “two eyes” a negative weight. Experiment with stuff like this to learn how it can help you.
third-eye, felinoid::1 two eyes::-0.3
Learn more on Discord: https://discord.com/channels/662267976984297473/1020578042285920327
In --v 5, write in sentence-like fragments, not word lists. “A young woman walks across campus” will work better than “woman, young, walking, campus”.
In --v 5, to generate something other than a photographic image, you will need to reference art movements, artistic techniques, genres, media type, game titles, directors, artist names, influences, time periods, etc.
To invoke the aesthetic style of an image, try referencing two or more of these:
My name is @whatnostop (clarinet) on the Midjourney official Discord server and I work for Midjourney. 🤓
We offer 24/7 community support and 100+ FAQs and tutorials for promptcraft, free with your membership.
Search by keyword on Discord #prompt-faqs or #prompt-chat forums for more help:
https://discord.com/channels/662267976984297473/1017917091606712430
Last updated by @whatnostop (clarinet) 6:00 PM CST 10/25/2023