I noticed the same and had to think about that scene in the Spaceballs movie where the guys face is rotated to his back after beaming…
1 Like
Really nice! Other than the fact that it could probably use a hint of noise on the skin (or, some kind of skin texture added back on there) to look more realistic that turned out well. Even if they’re young and have smooth skin, it should still have more skin texture than that produced. And, the clothing on the same plane as the heads is still soft. Good progress on these models, but more room for improvement in a subsequent dev stage. Hair, eyes, teeth = good!
1 Like
Out of curiosity, I tried coloring with FotorPea (usually gives pretty good results) and then editing in Gigapixel AI (I don’t have the Studio version installed yet). I didn’t care about the background (it needs improvement). The coloring doesn’t seem bad to me. But it’s hard to know what colors were actually there. I wish you a lot of patience and luck with the b/w photos! But in the end it is a nice reward and a glimpse into the past.
4 Likes
Thanks! At the moment I am not taking too much time, just having fun with the new (to me) Gemini toy. I already did most of these with MyHeritage a few years ago but colorization has improved since then.
You’re making it really hard for me to resist Comfy, once again. 
1 Like
Alibaba.
Did ask Gemini about it because i don’t understand why Alibaba does open its models.
Alibaba’s decision to release its AI diffusion models under the Apache 2.0 license is a strategic move that may pursue several goals, typical of large technology companies in the open source space.
The Apache 2.0 license is a very permissive open source license. This means it allows users to use, modify, distribute, and sell the model for virtually any purpose as part of proprietary software—often without the obligation to publish their own code changes or their own software based on it under an open-source license (no “copyleft”).
Possible reasons for choosing the Apache 2.0 license
- Promoting acceptance and the ecosystem
Wide use and integration: The permissive nature of the license lowers the barriers to entry for companies. They can integrate the models into their commercial products and services without hesitation and without having to disclose the source code of their own modifications. This promotes the rapid and widespread acceptance of Alibaba’s models.
- Competitive advantage and market positioning
Competition with other tech giants: By providing high-quality, freely usable models, Alibaba is positioning itself in competition with other major AI providers (such as Meta, Google, and OpenAI). A permissive license can be an attractive differentiator, especially compared to more restrictive licenses or closed models.
Setting standards: If the models are widely adopted, Alibaba can indirectly set technological standards and consolidate its market leadership in the field of diffusion models.
- Commercialization via cloud services (AliCloud)
Distribution model: Although the model itself is free, such releases often aim to increase demand for proprietary cloud services (AliCloud). Companies that operate or further develop the models on a large scale require significant computing power, which Alibaba can offer through its cloud infrastructure. The open-source model serves as a “lead generator” for the core business.
Service and support: Alibaba can continue to offer commercial support, professional services, and specialized, proprietary extensions or optimizations for the open-source models
- Community contributions and improvement
Bug fixing and optimization: By opening up the code, Alibaba receives free feedback and contributions from a global developer community. This speeds up the identification of bugs, security vulnerabilities, and the implementation of new features.
Transparency and trust: An open license builds trust in the technology, as users can review the code and understand how it works.
- Legal clarity (patents)
The Apache 2.0 license includes an explicit patent license. This is an important legal aspect that provides users with additional security, as they are allowed to use the underlying technology without fear of patent lawsuits from contributors (including Alibaba). This makes the license particularly attractive to businesses.
I think it provides a very good basis for rectifying these faults afterwards.
OK, just tried Qwen, thanks for the heads-up. In this case, Gemini was better and truer to the original:
Gemini:
Qwen:
1 Like
Definitely!
I’m just saying these models have come a long way. But there’s still some development headroom. Especially if those things can be handled in fewer steps. Not an issue for me b/c I don’t mind noodling around. But pros probably want to minimize time spent tinkering in order to maximize profits on jobs.
If the bodies can be masked (so the heads don’t get doubly fixed) - some of the AI sources currently being used/mentioned above might be able to correct the soft bodies (gee, that makes me feel like I should go to the gym as one example…) syndrome too!
The primary advantage of using Qwen is its ability to process images locally, eliminating the need to upload them to a cloud service and thus ensuring privacy. Furthermore, Qwen has no usage limits or quotas, offering unrestricted flexibility. When paired with ComfyUI, users can batch-process multiple images in a folder or generate a wide range of variations from a single image by applying different seeds.
3 Likes
Here is my attempt using Qwen ComfyUI. 
I didn’t want to hijack the Gigapixel post, so I shared my result here.
3 Likes
Very interpretive!
1 Like
Thanks Jackie!
1 Like
Very realistic colors.
1 Like
