A New One-Shot Fine-Tuning Notebook for RuDalle by kaliyuga

hive-158694 · @kaliyuga · May 23 '23 (edited)

$19.62

A New One-Shot Fine-Tuning Notebook for RuDalle

[Since I wrote about the release of RuDalle a couple months back](https://peakd.com/hive-158694/@kaliyuga/rudalle-a-fully-trained-accessible-multi-billion-parameter-answer-to-openais-dall-e-model-in-russian), RuDalle has taken the AI art community by storm. One major downside for me as a non-coder, though, was that there was no easy, idiot-proof way to train the RuDalle model on my own datasets or single images. Luckily, people who *can* actually write code have sort of solved this problem for me in the intervening months. Twitter user [AI_Curio](https://twitter.com/ai_curio) released a tool called Looking Glass AI in early December that allows you to fine-tune RuDalle on your own image (or images), and while it doesn't seem possbile to actually save the fine-tuned models at the moment (although that will hopefully be fixed in an upcoming notebook version), I've found it a really satisfying tool to learn and play around with. It's pretty self-explanatory, since AI_Curio took the time to include instructions in the notebook. 

Here's some examples of what I've been able to create using a dataset of 55 images. It successfully blended a number of input shapes and styles in really interesting ways and created something totally new! (They are not yet upscaled, so if they look slightly blurry, zoom out 😀)
![Screen Shot 2021-12-27 at 11.39.26 PM.png](https://files.peakd.com/file/peakd-hive/kaliyuga/242sBzKZkBcoG8jvah8Mj5GSUXTVVWz9GgGUXFk3U1dBb4e8yunjJUmnfz8rFWxh3gS1x.png)
![Screen Shot 2021-12-28 at 3.21.08 AM.png](https://files.peakd.com/file/peakd-hive/kaliyuga/23uFisHtatiTAMDVPa7QAd4Kdhe7y4gzwLZptMDPUHCvbQ6qxE4BaRBaoEhjYteCP4p64.png)
![Screen Shot 2021-12-28 at 2.13.27 AM.png](https://files.peakd.com/file/peakd-hive/kaliyuga/2423x4ohr5RCn9PYsFAu4kpJp8USac7onKD8fwhntEsbxdtdS3uZCwxUeLmLf4c5pV3mK.png)


--------
**A link to the current version of the notebook can be found [here.](https://colab.research.google.com/drive/15vFLeepkSTr1qd4xs31g9kMEiwkWP0sh?authuser=0#scrollTo=Vj_dK6U1nED3) Please note that there are copyright limitations to what you can use the outputs for.**

👍 pfunk, roelandp, juliakponsford, pishio, eturnerx, arcange, mmmmkkkk311, kevmcc, laruche, quixoticflux, clayboyn, vachemorte, nftshowroom, ctime, veeart, motionkapture777, voter003, steemitboard, rootdraws, limn, walterjay, mariaced, mariuszkarowski, icetea, fengchao, an-man, abstrads, peachymod, julesquirin, raphaelle, jdc, auditoryorgasms, nfthypesquad, shainemata, pboulet, informator, steempower-001

`author`	kaliyuga
`permlink`	a-new-one-shot-fine-tuning-notebook-for-rudalle
`category`	hive-158694
`json_metadata`	{"app":"peakd/2021.09.1","format":"markdown","tags":["art","aiart","painting","digitalart","generativeart","experimental","illustration"],"users":["kaliyuga"],"image":["https://files.peakd.com/file/peakd-hive/kaliyuga/242sBzKZkBcoG8jvah8Mj5GSUXTVVWz9GgGUXFk3U1dBb4e8yunjJUmnfz8rFWxh3gS1x.png","https://files.peakd.com/file/peakd-hive/kaliyuga/23uFisHtatiTAMDVPa7QAd4Kdhe7y4gzwLZptMDPUHCvbQ6qxE4BaRBaoEhjYteCP4p64.png","https://files.peakd.com/file/peakd-hive/kaliyuga/2423x4ohr5RCn9PYsFAu4kpJp8USac7onKD8fwhntEsbxdtdS3uZCwxUeLmLf4c5pV3mK.png"]}
`created`	2021-12-28 23:15:27
`last_update`	2023-05-23 16:42:54
`depth`	0
`children`	0
`last_payout`	2022-01-04 23:15:27
`cashout_time`	1969-12-31 23:59:59
`total_payout_value`	9.824 HBD
`curator_payout_value`	9.794 HBD
`pending_payout_value`	0.000 HBD
`promoted`	0.000 HBD
`body_length`	2,101
`author_reputation`	243,481,853,306,775
`root_title`	"A New One-Shot Fine-Tuning Notebook for RuDalle"
`beneficiaries`	`[]`
`max_accepted_payout`	1,000,000.000 HBD
`percent_hbd`	10,000
`post_id`	108,943,403
`net_rshares`	9,143,224,684,302
`author_curate_reward`	""

properties (23)vote details (37)

voter	rshares	pct
pfunk	7,322,477,009,866	100%
roelandp	423,955,830,098	10%
arcange	173,417,524,146	2%
raphaelle	1,273,040,423	2%
clayboyn	26,283,975,582	25%
walterjay	5,551,383,856	0.35%
steemitboard	6,571,349,401	2%
jdc	1,043,168,866	20%
vachemorte	21,090,243,925	25%
juliakponsford	316,437,223,573	50%
auditoryorgasms	722,446,165	25%
eturnerx	271,512,105,753	6.6%
kevmcc	43,238,240,040	30%
mmmmkkkk311	83,009,789,964	10%
mariuszkarowski	3,755,066,086	10%
informator	519,462,912	5%
pboulet	567,964,036	0.7%
voter003	6,788,621,212	3.5%
shainemata	588,529,555	1%
ctime	14,800,710,735	2.5%
nftshowroom	20,886,350,048	50%
peachymod	1,759,638,571	50%
icetea	2,797,381,280	60%
rootdraws	6,334,279,057	50%
steempower-001	438,481,613	10%
julesquirin	1,505,112,913	10%
fengchao	1,960,353,971	1%
laruche	30,828,420,024	0.7%
veeart	13,855,242,161	100%
mariaced	5,327,647,457	100%
an-man	1,888,807,293	100%
limn	6,194,152,949	21%
motionkapture777	13,108,571,316	100%
quixoticflux	29,562,746,478	100%
nfthypesquad	605,540,205	10%
abstrads	1,772,612,168	100%
pishio	280,795,660,604	5%