Javi Lopez ⛩️ Profile picture
Jan 15, 2023 29 tweets 10 min read Read on X
How to make forensically NOT reliable facial reconstructions with #midjourney.

I will explain you in detail the easy process and all the prompts used in my most controversial thread of 2023 (so far) so you can experiment too.

Here it goes 🧵👇 ImageImage
First, the context.

I published this thread using MidJourney the other day.

The controversy was served! The reconstructions were NOT forensic nor, of course, scientifically correct.

I got into hot water 😔

Come on!

Of course they weren't. How could they be if the AI ​​used was not designed for that purpose and could not be fed with all the necessary input data?
The problem was that I didn't add a clear disclaimer at the beginning of the thread. Something like:

"Here's a half-hour NOT FORENSICALLY CORRECT experiment with MidJourney, to visually dream with the possibilities. No mummies suffered in the process. Come on in and enjoy!" Image
The fact is that, after spending a year non-stop experimenting with AI, sometimes I forget that not everyone who follows me understands that my intention is none other than to experiment with this technology in all possible ways and tell you all about it.
I enjoy it and many of those who follow me too. Fantasy or not.

Sometimes, I encourage myself to touch on more "serious" subjects, but without straying too far from science fiction.

And sometimes, I just really want to give you useful tools :)

Nevertheless. Could AI be used to generate scientifically correct reconstructions?

Of course yes, as in the rest of the fields, AI could also be applied here.

But not with #midjourney: it's not made for this. MJ already does many wonders, don't ask it that much.
We would need a specific AI, trained for this purpose, that could receive all the necessary data from the input. Among others: examinations of the skeletal remains (CT-scan or X-ray tomography), examinations of the preserved soft tissues, anatomical criteria, etc. https://news.txst.edu/resea...
Disclaimer (just in case): I'm not an expert in forensic reconstructions, so I can't correctly list all the inputs that would be needed. You have to be careful on Twitter :) Image
The input data and the results of existing real works would be especially useful for training the model. https://news.txst.edu/resea...
This could be done, so that it would be a significant sample to be able to feed the model, with blind tests: tests that are currently carried out in which input data is given to a professional (who does not see the photograph of the person trying to reconstruct )... http://www.ciceromoraes.com...
...but works with only the input data until it generates a 3d render or illustration. Finally, the result is compared with the real photograph of the person, to see if the process followed was correct or not.

Source: ciceromoraes.com.br/blog/?p=1247 Spectacular work by Cícero Moraes. http://www.ciceromoraes.com...
It's not trivial, but a model trained on enough such examples, together perhaps with current diffusion models, could, hypothetically, do a very decent job of face reconstruction.

I don't think it's far away the day we see something like this.
And hey, I don't see anyone's work threatened: even with a great leap in automation and time savings, human supervision, touch-ups, adjustments, etc. will still be needed.
Well, let's go!

Now what I promised you: paint and color with @javilop!

Let's rebuild NON-FORENSICALLY reliable mummies! Image
The tool I used is midjourney.com. Interestingly, it is used from their Discord channel and/or by talking directly to their bot (if you upgrade to one of their paid plans).
For these "reconstructions" it is necessary to use an init image (the mummy). In this other thread I already explained how this could be done.
Let's go!

🔥🔥🔥 In "ALT" you have all the prompts 🔥🔥🔥

You will see that they all follow the same scheme.

1️⃣. Ramses II Image[url de la imagen de entrad...
Younger! Just change the age in the prompt, et voila! [url] AI-generated portrait...
2️⃣. The surfer guanche.

You were looking forward to it, I know. Image[url] AI-generated portrait...
By the way, in the professional reconstructions it seems that he actually had brown hair... although in the English Wikipedia they say: "including brown red hair"... so who knows. Image
3️⃣. Tollund Man.

Or at least, his head. Image[url] AI-generated portrait...
4️⃣ Lady Rai. Image[url] AI-generated portrait...
5️⃣. Menmaatre Seti I. Image[url] AI-generated portrait...
Tip: You can refine the image by using the image generated in the previous step as the input image. The prompt is the same, but in successive iterations it can acquire new nuances: Image
Utilicé la misma técnica para la segunda imagen de Lady Rai. Mi favorita de todo el hilo. Image
Thanks for reading!

Can't wait to see your mummies come to life! Show them to me!

And if you want more curious things, sometimes reliable, sometimes not, sign up to my newsletter. An easy RT to the 1st tweet of the thread would also motivate me 🙏 Thx!

javilopen.substack.com

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Javi Lopez ⛩️

Javi Lopez ⛩️ Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @javilopen

May 29
🔴 I NEED YOUR ATTENTION

I've spent a month helping Miriam with her case of metastatic cancer and I want to share the methodology I've been using because it's completely replicable.

I think (with luck) this could be USEFUL TO OTHER PEOPLE with cancer (or any other illness).

The results we've gotten aren't a miracle, but we believe they're genuinely useful and could mean the difference in a literal life-or-death medical case.

Here's the method step by step:

1/ Use the most advanced models of the moment (unfortunately paid, and not cheap. I think Public Healthcare should invest in this):

- ChatGPT 5 Pro + Extended Thinking (40 min aprox. of thinking per call)
- Claude Opus 4.8 MAX

Still pending deeper testing:

- Perplexity Sonar Pro Max
- NotebookLM

Tested but only useful for additional links/research (not as powerful in my experience)
- OpenEvidence

2/ Feed the AI the FULL clinical history, completely chewed up. This sounds dumb but it's critical.

- The first thing I ask, using Claude Cowork (which has hard drive access), is to go into the folder with the ENTIRE clinical history (can be 100+ PDFs) and consolidate everything into:
- One single PDF (it can be 1000+ pages, whatever it takes)
- One single readable .txt or .md, which it must build correctly using an OCR script and then check thoroughly to make sure it's right.

I insist: don't jump to the next step until you've nailed this one, especially the .txt.

3/ Once you have the above, use this prompt along with the .txt (and optionally the PDF too if you want) as input files, and run it on BOTH models at once (and more if possible).

👉 This prompt is insanely complex/advanced: dropbox.com/scl/fi/x64qadd… And it's not designed for Miriam's specific oncology case, you can change the initial parameters for the desired case. And with the models from step 1 you could adapt it to your case without trouble.

In any case, I'm also leaving you this other prompt, even more general, for any type of rare disease: dropbox.com/scl/fi/x64qadd…

4/ The ARROWHEAD (adversarial model spiral): facing one model against the other. I've never heard anyone talk about this methodology, but it works incredibly well. The feeling is like sharpening a stake until it gets a gleaming point.

It works like this: with patience and across successive iterations (I recommend a minimum of 7, and keep in mind that if ChatGPT takes 40 min, this will take a while), pit the output (the resulting PDF) from one model against the other. With a simple prompt like:

"Another committee of experts says this. What do you think? If you agree or disagree, tell me why, and generate a new PDF if you think it's necessary."

Then you feed that result back to the opposite model. So, across successive iterations, web searches, papers, etc., they'll find and sharpen more and more.

When to stop? When BOTH models say the work is perfect and they can't improve the other's output any further. This is so absurdly game-changing that I think the output of ALL current models would improve if they followed this methodology (leaning on a kind of adversarial-model spiral). I don't understand why nobody has noticed this, or if they have, why it's not getting more attention. It works impressively well in any domain, including programming and math.

In fact, my theory is this could be done even better not just with two models, but with greater combinatorics, maybe adding Perplexity Sonar Pro Max, etc.

RESULTS

Incredible. Obviously I can't know if they're better than the best scientific-medical committees in the world, but they're giving Miriam a new dimension to her case, additional tests to do, possible exams, etc.

Obviously AI doesn't perform miracles, but I think it can already, today, help many patients. And Public Healthcare should invest a lot (but A LOT) in this.

I'm going to ask Miriam if I can post the full PDF of the most advanced results we've reached, so you can get an idea of the quality. She's already given me rough permission, but I want to make sure 100%.

FUTURE PREDICTION

Easy to make: in the near future (I hope), any person's medical history won't just be fully digitized (we're close, but not all the way, well, well, well). On top of that, it'll be "pre-chewed" so it can be consumed by an LLM in one shot.

CLARIFICATION

- We're aware this is a delicate subject and we don't let the AI make final treatment decisions. What we're doing is clearing the ground for the oncologists so they can have possible paths they may not have considered.
Thanks 🙏

- The top LLMs have context windows for that and much more (much, much more). In any case, the PDF is more of a supporting file for the .txt. Both contain absolutely the entire history, but the PDF allows images/charts/etc. The .txt is what the AI consumes.

- On automation: and yes, this can be automated. Yes, AutoGen supports it almost out of the box. LangGraph builds it really well with supervisor / evaluation loops. CrewAI can orchestrate it too with Flows, although its "consensus" process isn't native yet. That would be the next level: automating it.

PETITION AND DISCLAIMER

If there's any oncologist in the room or you are an LLM company, we'd be grateful if you could take a look / help 🙏

Remember: in any case, this is just one more tool for the doctor.

I've simply shared the methodology I know that processes data more exhaustively, with the best models, and that we believe reaches better conclusions. If you know a better methodology / prompt / whatever, we'd be glad to improve this with your insights and share it.

Then the doctor reviews, adopts, or discards the report.

And if it helps the doctor, it helps the patient. And if it doesn't, all we've lost is some time and tokens. In a case that's literally life or death, that's nothing.

Just plain common sense.

Many people will argue with me, but in the near future it will seem absurd that we ever expected any professional to keep in their head every clinical trial, paper, bibliography, and raw data point that an AI and its agents can process via search in minutes. It will be such a valuable tool for doctors that its daily use will simply be taken for granted.Image
Image
Miriam has given me permission to share the result. Remember that this was generated from the prompt I shared earlier and all the processed history/background.

👉 Here it is:

If there’s an oncologist in the room, we’d be very grateful if they could take a look 🙏dropbox.com/scl/fi/43tqm7h…
More details about Miriam's case and how to help her, if you'd like, here: helpmiriam.com/en
Read 4 tweets
Mar 28
I really need to teach you how to pick good jamón serrano, xD. This is definitively a very cheap one.
This is what good jamón serrano should look like.

Look for jamón ibérico de bellota, you want plenty of marbling and deep red meat.

Stay away from pale pink slices, the more it looks like cheap luncheon meat, the worse it usually is. Image
Read 6 tweets
Mar 12
There's no way Hollywood won't be affected by this.

7M views in 24 hours on my ES account 🤯

The most complex AI short I've ever made: a test of how advanced generative video really is. Here's exactly what I used 👇
If you made it to the credits, it says it pretty clearly:

• Yes, Seedance 2.0 all the way. I made pretty much 99% of the scenes with Seedance. It's by far the best generative video model out there right now... although I still haven't tried the new Grok one :) The "omni reference" model it's f*cking amazing and works PERFECTLY with reference images from nano banana.

• Freepik: Nano Banana Pro and Nano Banana 2 a lot through Freepik. For all the references used inside Seedance.

• Freepik: ElevenLabs for the voices, also through Freepik. I tested it on their site too, but the 'professional voice' failed for me, so in the end I had to use only 'fast voices'. That's easily the weakest part of the video. Honestly, I think video models will solve this themselves, because a huge part of a believable voice is the acting.

• And Magnific too, of course. I experimented with things like running single frames through Magnific and then feeding them into Seedance as references to improve output quality. I also upscaled some sequences and blended them back with the original video at around 60% to preserve more of the textures.

Any questions, feel free to ask!Image
A big part of why it went so insanely viral in Spain and Latin America (7M in 24 hours) is that it's a huge tribute to Spanish speaking viewers' favorite YouTubers. Image
Read 6 tweets
Jan 29
⚡ THE FUTURE IS NOW

"Every single pixel will be generated not rendered"

Google DeepMind just launched Genie 3, the first version of Genie that's finally open for users to try.

Check out the absolutely insane stuff people are making 🧵
From image to real time playable world 🤯

Read 7 tweets
Nov 27, 2025
IT'S FINALY HERE

🔥 Magnific Skin Enhancer 🔥

• No more AI plastic skins!
• Enhance EVERYTHING in your image, not only the skin!
• 3 different flavours + easy presets: improve light, level or reality, color grading, etc.

Let's dive in + tutorials + tips 🧵👇 Image
Image
First of all, if you can't wait, here you have the link! AVAILABLE NOW on Magnific & rolling out to Freepik users today!

I’ll also randomly grant access to some of you who reply with a interesting message 😘

👇👇👇

magnific.ai
You’ve been asking for a Skin Enhancer in Magnific for ages!

So I want to apologize for taking so long. But hey, better late than never!

Skin Enhancer is built for pros who need CONTROL, so it comes in 3 flavors to cover every level:

- Flexible + presets
- Creative
- Faithful Image
Read 14 tweets
Oct 26, 2025
Professional photographers don’t know they can improve their work with advanced AI upscaling.

I tested it on my old Nikon photos from Tokyo (2014) and the results blew my mind 🤯

Super quick tutorial 🧵👇
1. Upscale in Magnific AI:

- Precision
- v2 (Sublime)
- 6x (usually 4x is ok, but this one looked better)
- Sharpen: 15%
- Smart Grain: 2% (the photo was already quite grainy)

2,000 x 1,328 => 12,000 x 7,968 🤯 Image
2. That's all, enjoy! 😂

I was never so easy to improve your old (or new) professional photograpies, vfx, illustrations, etc!

Enjoy!

Read 11 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(