avatarRenee LIN

Free AI web copilot to create summaries, insights and extended knowledge, download it at here

3875

Abstract

e=google&utm_medium=cpa_text&utm_campaign=ffxiv_evergreen_acquisition_anz_havas&utm_term=text1&gad=1&gclid=CjwKCAjwov6hBhBsEiwAvrvN6GfDrK-SRJHzk3JlFJ2DY0-Sta2ooX5ftrEYUiic0lWheT7HtR5hxxoCAYgQAvD_BwE&gclsrc=aw.ds">Final Fantasy 14</a></figcaption></figure><figure id="64be"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*8_QUNMJIul4HkUfNcqSuiA.png"><figcaption></figcaption></figure><p id="c281">Then use the URL at the beginning of the prompt.</p><figure id="8540"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*QL0tXh8SEfHSvBRa21gw5Q.png"><figcaption>Picture by author</figcaption></figure><p id="e58b">This is what I got in cyberpunk style; I like it!</p><figure id="7b3c"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*rlGPN7KMuQ5GU6SxTmZyfg.png"><figcaption>Picture by author</figcaption></figure><figure id="0e94"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*UP_a8w84i3cNRrd_pzAzfQ.png"><figcaption>Picture by author</figcaption></figure><h1 id="06f8">3. Combined the above functions</h1><p id="9831">Well, combining reference and seed functions is easy. I used the referenced cyberpunk generation above to make variations based on the “seed 101” I assigned. Check the beachside setting again.</p><figure id="22bf"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*cod7Sn6_LGnybndUrN9ulQ.png"><figcaption>Change the setting to beachside again.</figcaption></figure><p id="e20e">Now she is on the beach rather than in the city.</p><h1 id="fb8d">4. “describe” function and the anime model— “Niji”</h1><p id="0aac"><b>(1) The “describe” function</b></p><p id="9372">I have tried the new describe function; it can provide some keywords. But when you feed those descriptions back to Midjourney, it won’t get you images similar to the original as the “reference” function used above.</p><p id="d090">Just type “/describe” and paste the URL obtained before as in the first section above. I used the first and last descriptions to generate new images.</p><figure id="3786"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*i-yJQ93s33h72Etjcdb4hg.png"><figcaption>Picture by author</figcaption></figure><figure id="2abf"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*-NBLSoSDnCLCchkBIEo4DQ.png"><figcaption>The first description generated a male elf</figcaption></figure><figure id="cc57"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*FROD9gFLL7g_tkdZFxEvIQ.png"><figcaption>The last is better, but also a general look with long hair</figcaption></figure><p id="41c1"><b>(2) Niji model</b></p><p id="aefe">If you are an anime lover, you will love this specific model trained from animations. The model is a collaboration between Midjourney and <a href="https://spellbrush.com/">Spellbrush</a> tuned to produce anime and illustrative styles.</p><figure id="0a4b"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*qhYG-RHfo9XFLGmaHiwatA.png"><figcaption><a href="https://docs.midjourney.com/docs/model-versions">https://docs.midjourney.com/docs/model-versions</a></figcaption></figure><figure id="49ea"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*uG_u0pId99tIZ0GIZXU65w.png"><figcaption>The newest Niji 5</figcaption></figure><p id="f4dc">After using Niji, the cyberpunk style will change from the below Picture #1 to Picture #2</p><figure id="9217"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*p4wSnFbUo4L1fR9-YXvXXg.png"><figcaption>Picture #1</figcaption></figure><figure id="9059"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*bAN-LMRUvGn44az-dXFBkw.png"><figcaption>Picture #2</figcaption></figure><p id="870f">Further applying “style cute” and “style expressive” to the generation:</p><figure id="b439"><img src="https://c

Options

dn-images-1.readmedium.com/v2/resize:fit:800/1*w3L_HbTyqS-GNAQsFrcLCw.png"><figcaption>- niji 5 — style cute</figcaption></figure><p id="d8a6">I accidentally added both “cute” and “expressive” styles; the bot ignored the latter and gave the same result.</p><figure id="e1b7"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*Rq_1ae7ldk1K6-x4XHWaHg.png"><figcaption>Feed both style by mistake</figcaption></figure><p id="8bed">Corrected it to “expressive” style. Well, I like the cute style better.</p><figure id="14d6"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*PZuXo2Lxf8H79WDng_7FEQ.png"><figcaption>- niji 5 — style expressive</figcaption></figure><p id="893b">I upscaled the 4th one of the cute generations as a younger version of her.</p><figure id="5d38"><img src="https://cdn-images-1.readmedium.com/v2/resize:fit:800/1*OB_qgRh4mlXyWMcpdhMnGA.png"><figcaption>Picture by author</figcaption></figure><p id="3f74">Looking forward to using <a href="https://reneelin2019.medium.com/adobe-firefly-is-all-you-need-e166398a787c">Adobe Firefly</a>, and changing her eye colour, ear, etc, through text prompts.</p><p id="a520">If you liked posts like this, you might also like a <a href="https://reneelin2019.medium.com/membership"><b>Medium</b></a><b> </b>membership. It’s only $5 a month, but it will give you unlimited access to articles while supporting your favourite writers. If you sign up using this <a href="https://reneelin2019.medium.com/membership"><b>link</b></a><b>,</b> I’ll earn a small commission. Thanks!</p><p id="f5ed"><b>Here is the<a href="https://readmedium.com/table-of-contents-for-my-medium-posts-35fd72d6d27a"> Table of Contents for My Medium Posts</a></b></p><p id="7c4d">The other related posts on exploring how to use AI tools:</p><ol><li><a href="https://reneelin2019.medium.com/how-to-use-dall-e-2-to-create-images-from-text-17e041f81614">How to use DALL-E 2 to create images from text</a></li><li><a href="https://reneelin2019.medium.com/how-to-use-stable-diffusion-to-create-images-from-text-bbbc8a419f9">How to use Stable-Diffusion to create images from text</a></li><li><a href="https://reneelin2019.medium.com/how-to-use-dreamfusion-to-generate-3d-models-from-text-d5eb6d37a1c7">Using DreamFusion to Generate 3D Models from Text</a></li><li><a href="https://reneelin2019.medium.com/the-new-impressive-text-to-music-model-by-google-4c29a337638d">The New Impressive Text-to-Music Model by Google</a></li><li><a href="https://reneelin2019.medium.com/safe-ways-to-use-chatgpt-to-study-better-no-cheating-c1712d61c123">Safe Ways to Use ChatGPT to Study Better (No Cheating)</a></li><li><a href="https://reneelin2019.medium.com/four-problems-bother-me-when-using-midjourney-chatgpt-as-a-beginner-9f302fe328a7">Four Problems Bother Me When Using MidJourney+ChatGPT as a Beginner</a></li><li><a href="https://reneelin2019.medium.com/try-hugging-faces-open-source-text-to-video-model-modelscope-text-to-video-synthesis-9942a189e5d1">Try Hugging Face’s Open Source Text-to-Video Model (modelscope-text-to-video-synthesis)</a></li><li><a href="https://reneelin2019.medium.com/adobe-firefly-is-all-you-need-e166398a787c">Adobe Firefly is All You Need</a></li><li><a href="https://reneelin2019.medium.com/how-to-use-chatgpt-to-accelerate-literature-review-d14bfc47968b">How to Use ChatGPT to Accelerate Literature Review</a></li><li><a href="https://reneelin2019.medium.com/the-ultimate-midjourney-prompt-guide-2023-814770a98473"><b>The Ultimate Midjourney Prompt Guide 2023</b></a></li><li><a href="https://readmedium.com/what-is-prompt-engineering-will-it-last-long-do-we-need-to-learn-it-1303f4e04ad4">What is Prompt Engineering? Will it Last Long? Do We Need to Learn It?</a></li><li><a href="https://readmedium.com/chatgpt-in-healthcare-fef2ccc788d9"><b>ChatGPT in Healthcare</b></a></li></ol></article></body>

Creating Consistent Characters with Your Desired Reference Characters/Pictures in Midjourney

Picture by author, Alisaie from Final Fantasy 14

I believe that creating consistent characters in multiple settings is quite helpful. Besides that, most of the time, we want to create characters similar to those we like. Therefore, I wanted to know how to have these kinds of better control over the generations at the beginning(Four Problems Bother Me When Using MidJourney as a Beginner ). I explored Midjourney’s documentation in this post and considered combining the “seed” and “reference image” functions to suit my needs.

  1. Consistent: fixing the “seed” (which fixes the initial settings of the neural networks)
  2. Reference: upload and use the image as a reference
  3. Combined the above functions
  4. The worth noting “describe” function(new) and the anime model — “niji”

1. Consistent: fixing the “seed”

First, find a generation created two days ago; find its seed. You could click the inbox in the top right corner and jump to that generation. It’s worth noting, Midjourney knows “Splatoon”.

You can add “ — seed [number]” when you create the image; in that way, you won’t need to find it every time. Details are explained in this post.

Find a generation you created before

Send the “envolope” emote:

Picture by author

Then, the bot will send you the info, which you can find under the “Unreads”

Send the “envolope” emote

Use the same prompt with the seed, but change the setting to the beachside. They are the same characters as above.

Using seeds

If you don’t use the seed, the generation will be less similar:

Without the seed

You can change their clothes to kimono.

Keep adjusting the setting

2. Reference: upload and use the image as a reference

Upload the image:

Click the image, and open it in the browser to find the URL:

Alisaie from Final Fantasy 14

Then use the URL at the beginning of the prompt.

Picture by author

This is what I got in cyberpunk style; I like it!

Picture by author
Picture by author

3. Combined the above functions

Well, combining reference and seed functions is easy. I used the referenced cyberpunk generation above to make variations based on the “seed 101” I assigned. Check the beachside setting again.

Change the setting to beachside again.

Now she is on the beach rather than in the city.

4. “describe” function and the anime model— “Niji”

(1) The “describe” function

I have tried the new describe function; it can provide some keywords. But when you feed those descriptions back to Midjourney, it won’t get you images similar to the original as the “reference” function used above.

Just type “/describe” and paste the URL obtained before as in the first section above. I used the first and last descriptions to generate new images.

Picture by author
The first description generated a male elf
The last is better, but also a general look with long hair

(2) Niji model

If you are an anime lover, you will love this specific model trained from animations. The model is a collaboration between Midjourney and Spellbrush tuned to produce anime and illustrative styles.

https://docs.midjourney.com/docs/model-versions
The newest Niji 5

After using Niji, the cyberpunk style will change from the below Picture #1 to Picture #2

Picture #1
Picture #2

Further applying “style cute” and “style expressive” to the generation:

- niji 5 — style cute

I accidentally added both “cute” and “expressive” styles; the bot ignored the latter and gave the same result.

Feed both style by mistake

Corrected it to “expressive” style. Well, I like the cute style better.

- niji 5 — style expressive

I upscaled the 4th one of the cute generations as a younger version of her.

Picture by author

Looking forward to using Adobe Firefly, and changing her eye colour, ear, etc, through text prompts.

If you liked posts like this, you might also like a Medium membership. It’s only $5 a month, but it will give you unlimited access to articles while supporting your favourite writers. If you sign up using this link, I’ll earn a small commission. Thanks!

Here is the Table of Contents for My Medium Posts

The other related posts on exploring how to use AI tools:

  1. How to use DALL-E 2 to create images from text
  2. How to use Stable-Diffusion to create images from text
  3. Using DreamFusion to Generate 3D Models from Text
  4. The New Impressive Text-to-Music Model by Google
  5. Safe Ways to Use ChatGPT to Study Better (No Cheating)
  6. Four Problems Bother Me When Using MidJourney+ChatGPT as a Beginner
  7. Try Hugging Face’s Open Source Text-to-Video Model (modelscope-text-to-video-synthesis)
  8. Adobe Firefly is All You Need
  9. How to Use ChatGPT to Accelerate Literature Review
  10. The Ultimate Midjourney Prompt Guide 2023
  11. What is Prompt Engineering? Will it Last Long? Do We Need to Learn It?
  12. ChatGPT in Healthcare
Ai Art
Midjourney
AI
Animation
Generative Ai
Recommended from ReadMedium