For citation-. Feng Junlong, Long Junhao. Design and implementation of Al-assisted painting tools based on MidJourney // Grand Altai Research & Education — Issue 1 (21)'2024 (DOI: 10.25712/ASTU.2410-485X.2024.01) — EDN. https://elibrary.ru/GHGFMI
UDK 004.89
Design and implementation of AI-assisted painting tools based on MidJourney
Feng Junlong1*, Long Junhao2
1 Hubei Digital Textile Equipment Key Laboratory, Wuhan Textile University, Wuhan, 430073, China E-mail: *2419785330@qq.com ; 2889373053@qq.com
Abstract. With the rapid development of Internet technology, a large number of AI tools have emerged. Among them, the field of painting is the focus of people's attention. Through AI painting tools, people can simplify the painting process, which greatly facilitates people of different levels to create in the field of painting. However, the use of MidJourney is very complicated and consumes users' time. To solve this problem, this paper proposes a new architecture. Based on the Spring Boot architecture, it realizes interaction with MidJourney's official painting tool, combines the database advantages of Redis and MySQL, and introduces the message middleware RabbitMQ to achieve asynchronous communication and task decoupling within the system. The entire system aims to provide users with a stable and efficient platform for obtaining painting resources and promote innovation in the field of painting creation. The SparkDesk interface has been added to the system to realize the universal question and answer function of large models. The CRISPE framework is used to optimize the prompts used in the MidJourney and Spark model interfaces to further optimize the user experience.
Keywords: Spring Boot; Redis; Spark; painting field; MidJourney; CRISPE
MidJourney ЕЙ AI
шж*, жтт
i йж, 430073
E-mail: *2419785330@qq.com ; 2889373053@qq.com
ШШ: МЗДШЙ^Й'Ш^М, ХЙЙ AI ХШШ,
ЖЙ ai ^ихд, л т&шшжт, хх^шт^пшжй
Ш MidJourney шшти.
ЖЧШ, Spring Boot ШШ,
MidJourney ^7 Redis ^ MySQL ЙШШШ^Щ, ЩЛ
RabbitMQ, Itliii^
^t^AT SparkDesk gP, ^ITAfiMAmÎ^fô^m, CRISPE M MMt MidJourney ^ Spark flMf^P^fèfflfô Prompt, m-^ftffl^fà^.
^^M: Spring Boot; Redis; Spark; MidJourney; CRISPE
03m
MfêA^A^M, MidJourney ft
ftmm^^mx^MM. îM^AI^A®«^^
MidJourney É^ftx«, ^ffl^m^M^M^^^M^ ai m-^
» AI MM^AfêA AI frJftÉ^mftrntA.
1 Ai^mm^Mii^
MidJourney (GAN) m
rmiMÉ^ffl. gan A^É^Î^A
[2]. ^SA^AM
T, MidJourney g&iit,
a, ^^msA^mn® 1
«IS Stable Diffusion ^^fflA^A®,
Stable Diffusion Dream booth JA
Lora Textual Inversion JJA
Dream Booth É^ffifi^AM^, Lora ^A, M Textual Inversion l^ffifi^AÎÊ, JJUi^lM^WM^fJA^M [3]. Stable Diffusion ^S^ Text Encoder &
amA Noise Predictor m^iA?«, s^^ffl vae m^^fé, A^M^fô
SA, ÊAÎIAim (stable
diffusion) GPU AëÎT,
fJTAr±№ffl [4].
Ш1 шшшш
Figure 1 System function module diagram
ШршшШ, яр^-щыш ai тя^нтп ai ^ИШ. Ж+ШРЯ
шмш. ЯЯШ^Ш^Ж, ai ^И
ai ШШШШЯ
штшш.
(1) SparkDesk
l^T^ SparkDesk ШШШМШ. Ш2
1' 1 г U
ш щди А1ЩР
®2 SparkDesk ШРЯШ&ЙШ
Figure 2 SparkDesk chat module function diagram £ SparkDesk ЛШИ^ШЯШ*, ^Я^ФШШ^ШЙЛёВЙ«^
M ai тшшш. Шший вшт, Ш^^^ШМ^ШЙШ. ВДЯШРЯ ШШ&^&ШШШЖ, т^тштшшшлд. тяшшш
OpenAI ЬШШ.
(2) MidJourney ШШШШ
MidJourney ^ттштт. адз fjpk
Aims
1 1 1
sa и hse и mm №ср
MidJourney ^nm&Mrn Figure 3. MidJourney drawing module function diagram
2 Mid-Prompt ffiftr^^W
^^mmmx^k^m, MIdJourney
nm token, ^mm^rnrn^mmmm, m
Prompt Bt, M4 m^. ^WMM,
MidJourney
®4 ^Prompt Figure 4. Chinese Prompt
fêffl^f'Jft Prompt 0t, Ш5 ЯЯА. АШИШ^А
Figure 5. Highly thermally conductive, high-strength graphene-based composite films
by eliminating nanosheet wrinkles
ai ААИ^ШАЯА^п^Ат
ШАйёВШАШШЖ ЯШ MidJourney ö Prompt i
a^, ША, ад, швд, ш, ш, ш.
АШПШШФ^ЗД Prompt 0А
£пТ®б ЯЯА
Figure б. Use structured prompts
3 Prompt ffift-CRISPE m^&fr
Prompt iMKMlM, MM^iiiffl Prompt ffifofofi CRISPE ffl^UM^^M
flM^TS^. CRISPE CRISPE m^:
CR: Capacity and Role (iA^fe). AI
I: Insight (»), S: Statement (P№), AI
P: Personality (^ft), AI Uf^M ^A^M^i.
E: Experiment AI Aili^^M.
»a Mm^m^rnmrnw^ti
mm&amm&m. Mrasmidjoumey Aigai«. ttM—ssE, smmrnsm
m-^mn&om, nmmsm-Tm^imm, tmn-msm, ^eissMsss, Jta^^a-^ssi®,
mSiJUi^'J-CRISPE mmzft
Figure 6. Dialogue training example-CRISPE framework design
4 mi&m
^M^^ffl Java spring boot feA^^, iiffi MidJourney-proxy f^fMM^P AM. MSGitHub chat next web M»AM. M^tiffl№
AI ^WIMAMM MidJourney image MidJourney ff
^7 Discord Bot feA, Discord
Stable Diffusion m OpenAI № AI
^t^A Stable Diffusion m OpenAI № AI
2021. DOI:10.27239/d.cnki.gnhhu.2020.000814.
[2] GAN g^fcfffi, 2017, 43(03):321-332.D0I:10.16383/j.aas.2017.y000003.
[3] Mokady, Ron, et al. "Null-text Inversion for Editing Real Images using Guided Diffusion Models." ArXiv (2022).
[4] mxrn. ai ^bma g^efj&wtim. 2023,
5(03):424-430.
[5] Appleone. (2023-12-05). Prompt ffiA https://zhuanlan.zhihu.eom/p/670552915.
[6] AI ^i^tSft. (^ AI ^fM). https://zhuanlan.zhihu.eom/p/677432362?utm_psn=1762030859241132032.
References
[1] Wang Chenwei. Comparative Research on "Artificial Intelligence Painting" and Artist Painting from the Perspective of Creative Theory [D]. Nanjing University, 2021. DOI: 10.27239/D.cnki.gnhhu.2020. 000814.
[2] Wang Kunfeng, Gou Chao, Duan Yanjie, et al. Research progress and prospects of generative adversarial network GAN [J]. Acta Automata, 2017, 43(03): 321-332. DOI: 10.16383/j.aas.2017.y000003.
[3] Mokady, Ron, et al. "Null-text Inversion for Editing Real Images using Guided Diffusion Models." ArXiv (2022).
[4] Ma Yuan. The Future of AI Painting from the Perspective of Jurisprudence: The Balance of Freedom and Control [J]. Journal of Intelligent Science and Technology, 2023, 5 (03): 424-430.
[5] Appleone. (2023-12-05). Prompt optimization https://zhuanlan.zhihu.com/p/670552915.
[6] Shangjin AI painting creative base. These tools are awesome! Generate Midjourney prompt words with one click! (AI tutorial attached). https://zhuanlan.zhihu.com/p/677432362?utm_psn=1762030859241132032.